You are on page 1of 234

ON ECONOMIZING THE THEORY OF A-BAR DEPENDENCIES

by

WEI-TIEN TSAI

B.A. National Ts'ng Hua University


(1986)

M.A. National Tsing Hua University


(1989)

Submitted to the Department of Linguistics and Philosophy


in Partial Fulfillment of the Requirements for
the Degree of

Doctor of Philosophy

at the

MASSACHUSETTS INSTITUTE OF TECHNOLOGY

September 1994

@ Wei-Tien Tsai

The Author hereby grants to MIT permission to reproduce and


to distribute copies of this thesis document in whole or in print

Signature of Author
Department of Lingujisticsa.d Philosophy
J ,d2It994
Certified by
Noam Cho sky
Instjtute Professor

Accepted by
Wayne O'Neil
Head, Department of Linguistics and Philosophy

ARCHIVES
MASSAr.HUSUTfrc IJSTITUTF

DFE 21 1994
OnEconomizing theTheoryofA-BarDependencies
by
Wei-Tien Tsai
Submitted to the Department of Linguistics and Philosophy
in Partial Fulfillment of the Requirements
of the Degree of Doctor of Philosophy
Abstract
This dissertation aims to derive linguistic variations of wh-question
and syntactic asymmetries among wh-expressions from a fairly restricted
set of factors under the minimalist approach: 1) binary vs. singulary
substitution (i.e., Generalized Transformation vs. Chain formation), 2)
noun vs. adverb, and 3) weak vs. strong operator features. Correlations
have been established between wh-question formation and quantification in
terms of the structural height of binders, as well as the magnitude of
unselective binding. Chinese, English, Hindi, and Japanese are examined
to give a selective but representative spectrum of this correlationship.
On empirical grounds, we demonstrate that unselective binding and
(A'-)Chain formation are different breeds of construal. It is shown that the
asymmetries between unselective binding and long wh-movement in
general reflect the distinction between binary and singulary substitution.
Our second goal is to relate the (in)definiteness/specificity of nominals
to their structural properties. By extending Diesing's (1992) mapping
hypothesis, we present a fairly explicit mechanism of mapping syntactic
representations to their corresponding logical forms, centering on the
notion of syntactic predicate. This move provides us a simple and optimal
way to characterize the interaction between predication and quantification.
Existential closure is also shown to observe the Greed principle if
understood properly, i.e., as an interpretation procedure rather than a
syntactic operation. We also explore the possibility of eliminating the
lowering mechanism in favor of the copy theory, and initiate an attempt to
reduce the stage-individual asymmetries to the distinction between degree
and individual variables in the sense of Helm (1987) and Frampton (1990).

Thesis Supervisor: Noam Chomsky


Title: Institute Professor
ACKNOWLEDGEMENTS

Being an outlander, I often feel awkward in expressing either regret or


gratitude, sometimes too little, sometimes too much, sometimes untimely,
sometimes timely but ungracefully. Since here I need to deal with only one of
them (and the easier one, too), I hope I can do a better job.
Above all, I am in great debt to the members of my dissertation committee:
Noam Chomsky, Ken Hale, James Huang, and Howard Lasnik. From the first
time I picked up a linguistics textbook to the end of my thesis defence, Noam
has always been the inspiration to me. It's a long, long journey, but I never
regret it. I would like to thank him for sharing his visions with us, and for putting
up with my gibberish and ignorance for the past four years. From Ken, I learned
the self-devotion of a humanist, the sage of a hermit, and the chivalry of a lone
ranger. More than once, he drew me back from the brink of being devoured by
technicality and shortsightedness, and I often suspect that he is the only person
who actually has the UG. Since the early years of my career, Jim Huang has led
me to appreciate the art of argumentation and the beauty of the generative
architecture. He gives me a sense of direction in.this intellectual maelstrom,
making me proud of being one of the Chinese linguists. Most importantly, he
has helped me to find out who I am, and where my root is planted in. Howard,
on the other hand, gives me a sense of history and tradition, alarming me of the
everlasting arrogance residing in intellectual minds. He also shows me in
person the definition of open-mindedness, and the power of logical reasoning.
I would also like to thank the members of my generals committee: Morris
Halle, Wayne O'Neil, and David Pesetsky. Morris used to exclaim to me in his
vigorous voice "We want to educate you!", which he did. He rid me of self-pity,
and kept me going through those darkest hours of my life. Wayne, being a
caring educator, has listened to my problems with endless patience, never
reluctant to offer helps. David has been extremely helpful either as a brilliant
lecturer in class, or as an observing critics during our appointments. I thank him

3
for trying hard to make senses out of my self-indulgent narration and not so
illustrious illustrations.
I am equally grateful to Irene Heim, James Higginbotham, Michael
Kenstowicz, Alec Marantz, and Shigeru Miyagawa. As a complete stranger of
formal semantics, I have been picking up my meager training in Irene's classes.
Her meticulously prepared handouts have become legendary (for that matter,
we should thank Angelika Kratzer as well, who coauthors the handouts). In spite
of my slowness and syntax-centricism, she has always been kind and ready to
help, particularly in providing invaluable insights into the issues of existential
closure and syntax-semantics mapping. Jim Higginbotham is one of a handful
of people who can play the double role as a syntactician-semanticist, and plays
it well. He matches my stereotypical image of a "professor" most perfectly, and is
inspiring in every way. It is Jim and Irene who convinced me the necessity of
converging syntax and semantics. I learned from Michael not only the cream of
recent accomplishments in phonology, but also his methodological approach of
problem-solving. I think anyone around Building 20 would agree that Alec has
made himself available as often as one can get. I thank him for disentangling
my impossible reasoning thread by thread, and for the disillusion of my self-
assertion, now a nostalgic remembrance of my reckless youth. Shigeru has
always been encouraging and supportive. He is not only a wonderful teacher,
but also a caring friend.
Furthermore, I would like to express my admiration of Lisa Cheng, Audrey
Li, and Jane Tang, whose influences will be evident as our plot unfolds. When I
first came to Boston, Lisa took me, an "intellectual brother" in Chinese terms,
under her wings. Since then, I have been benefiting tremendously from her
example, both in terms of academics and real life. I thank her for the guidance
and timeless friendship. Audrey has pioneered the research of several major
issues of this thesis. We agree and disagree. But we always share the same
intuition as to which direction we should proceed and which methodology we
should be equipped with. I thank her for seeing things in wonder eyes. For the
past four years, Jane has been an inexhaustible source of comforting and
encouragement. She is not only constructive in comments, but also resourceful
in suggestions. I thank her for being a devoted friend and a selfless comrade.
I also owe this career to my teachers back home at the National Tsing Hua
University: Peter Cole, Tom Ernst, Hsin-I Hsieh, Chu-Ren Huang, Jen-Kuei Li,
Chin-Fa Lien, Kuang Mei, Tsu-Lin Mei, Stanley Starosta, Tian-Xin Sun, Ting-
Chi Tang, Pang-Hsin Ting, Feng-Fu Tsao, William Wang, Samuel Wang,
Guang-Yu Zhang, Yue-Qin Zhang, to my old associates: Ying Zheng, Hui-
Chuan Hsu, Fu-Wen Lin, Tzong-Hong Lin, Chen-Sheng Liu, Ching-Ching L"i,
Zhao-Ming "Big Brother" Gao, Jin-Wan Guo, Ming-Wei Guo, J. J. Li, Jo-Wang
Lin, Yi-Ching Su, Yung-Li Zhang, Wen Xu, Yu-Ling Xu, and to the new friends I
made here: Kathleen Ahrens, Zhi-Ming Bao, Dong-Dong Chen, Matthew Chen,
Su-l Chen, Robert Cheng, Bonnie Chiu, Raung-Fu Chung, San Duanmu,
Teresa Griffith, Yang Gu, Miao-Ling Hsieh, Ruoh-Mei Hsieh, Shi-Zhe Huang,
Thomas Lee, Ya-Fei Li, Feng-Hsi Liu, Sui-Sang Mok, Chun-Yan Ning, Ding-Xu
Shi, Feng Shi, Shu-lng Shyu, Chao-Fen Sun, Kuo-Ming Sung, Li-Mei Sung,
Rint Sybesma, James Tai, Sze-Wing Tang, Hong-Yin Tao, Jen Ting, Paul
Waltraud, Zhi-Qun Xing, Fei Xu, Jie Xu, Lie-Jiong Xu, Felicia Yu, Eric Zee,
Hong-Ming Zhang, Min Zhang, Ke Zou.
I am eternally grateful to Masa Koizumi, Masayuki Oishi, Asako Uchibori,
and Hiro Ura for their support in the process of completing this thesis. I will
never forget our little tea party, the feast at Cape Cod, and the incident of
"German cake". I too benefit greatly from discussions with John Frampton,
Sarah "Brenda" Kennelly, and Tanya Reinhart during their visits at MIT. I thank
John for his penetrating insights. Brenda has always been there for me when
things got tough, whose friendship I will cherish for life. Tanya's works has been
the inspiration of numerous proposals of this thesis. I am forever stunned by her
elegance in presentation and tenacity in argumentation.
Moreover, I would like to say thanks to Jun Abe, Andrew Carnie, Danny
Fox, Naoki Fukui, Elly van Gelderen, Sabine latridou, Alessandro Lenci, Marta
Lujan, Martha McGinnis, Anoop Mahajan, Orin Percus, Toshi Oka, Colin
Phillips, Norvin Richards, Uli Sauerland, Yuji Takano, Hubert Truckenbrodt,

5
Sarma Vaijayanthi, Akira Watanabe for their invaluable comments in various
occasions, and to Chris Collins, Hamida Demirdash, Koji Fujita, Tom Green,
Heidi Harley, Doug Jones, Yasuhiko Kato, Utpal Lahiri, Paul Law, Kumiko
Murasugi, Renate Musan, Pierre Picard, Jason Stanley, Hiroaki Tada, Chris
Tancredi for their friendship.
Special thanks to Jim Harris for his understanding and encouragement.
Baby-sitting my class must have been giving him a lot of headaches. I thank
Allen Katherine, Rachel Pearl and Jamie Young for their patience and kindness
in arranging the seemingly impossible defence date.
Finally, my most sincere gratitude goes to my classmates: Pilar Barbosa,
Jonathan Bobaljik, Tony Bures, Diana Cresti, and Seth Minkoff From the first
day we were summoned to do the registration, whi.., reminds me of a
kindergarten assembly, our destiny are bound together. In those bittersweet
days, we protested and we conspired, boasting about Bresnan's piano in our
office and offering each other a shoulder to cry on. Of course we argued. Over
smoking or non-smoking. Over astrology or astronomy. Still, nothing sticks but
our friendship and mutual appreciation. I thank them for making MIT a second
home for me, and for the fun and craziness throughout all these years.

6
TABLE OF CONTENTS

A bstract .... ........................................................................ 2


Acknowledgem ents ............................................................ 3
Table of Contents ................................. 7

CHAPTER 1 FROM LEXICON TO IF

0. To Move or Not to Move -- That is the Question ............... 9


1. Toward a Minimalist Design of Wh-Dependencies ........... 13
2. Singulary vs. Binary Substitution ................................ . 17
2.1. Disintegrate Lexical Integrity .................................... 17
2.2. Dou, Dou, Dou, and Dou ........................................... 23
2.3. Interrogative Construals in Chinese Bare
Conditionals .................................. 27
2.4. Japanese Mo-Quantification and Chinese
Dou-Quantification ..................................... . 37
2.5. A Conceptual Problem and an Empirical Solution ....... 48
3. Nouns vs. Adverbs .......................................................... 54
4. Strong vs. Weak Operator Features .................................... 65
5. An Overview ..................................... 73

CHAPTER 2 SOME ASYMMETRIES BETWEEN CHAIN


FORMATION AND UNSELECTIVE BINDING

1. Long-Distance Construals of Amount Wh 's ...................... 76


2. Interrogative Construals in Dou- and Mo-Quantification ...... 94
........ .... .. . . . . . .. .... 101
3. A'-Bound Pro or Pure Variable? .
4. Wh-Extraciton from Drived Nominals ................................ 107
5. Reflection on a Syntax-LF ntsymmetry .......................... 110
6. Aftermath .................................. ............... . 122

CHAPTER 3 TOWARD LF INTERFACE

0. 3-ClosureExtensions .................................... 123

7
1. Mapping as a Cyclic Operation .......................................... 129

1.1. The Cyclic Hypothesis .......................................... 129


1.2. The Mapping Geometry of Topics .......................... 135
1.3. Sum m ary ............................................................ 140

2. 3-closure as a Post-Cyclic Operation ............................ 141


2.1. The Nature of 3-closure ....................................... 141
2.2. Non-restrictive Relative Clauses .............................. 148
2.3. Existential Constructions and Predication Licensing ... 158

2.3.1. Existential Predicates and (In)definiteness


Restrictions .............................................. 159
2.3.2. Weak Existential Predicates aLd Secondary
Predication ............................................... 164
2.3.3. A Dilemma ................................................ 171
2.3.4. Bare Indefinites ............................................ 176
2.4. A Refinement .................................................... 178
2.5. Not Unlikely Extensions .......................................... 188
2.5.1. Small Clauses ........................................... 188
2.5.2. Secondary Temporal Predicates ....................... ... 190
2.5.3. Resultative Complements ................................ 198
2.5.4. A Preverbal-Postverbal Asymmetry of
Temporal Adjuncts ........................................ 200
3. Chain-formation as a Copying Operation ..................... 203
3.1. A Few Good Questions ......................................... 203
3.2. Lowering or Copying? .............. . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
3.3. An Individual Variable Account of Stage-Individual
Asym m etries
........................................................ 211
3.4. Disagree Chinese Agreement ................................. 218

4. Concluding Remarks .................................................. 226

References ....................................................................... 227

8
CHAPTER ONE

FROM LEXICON TO LF

1. To Move or Not to Move -- That is the Question

The past decade has seen a persistent evolution toward a minimal design
of grammar, or the theory of grammar to the same effect, within the principles-
and-parameters framework. A part of the intuitive content of this "minimalist"
approach, as proclaimed by Chomsky (1992), is to take interface levels to be
the only levels involved in derivation of linguistic expressions, with the notion of
"interface" strictly defined within the domain of (virtual) conceptual necessity. It
is suggested that the only levels needed are PF and LF, which serve as the
doorways from linguistics proper (including nothing more than a lexicon and a
computational system) to the articulatory-perceptual and conceptual-intentional
systems respectively. The bottom line is that the interface conditions on PF and
LF are satisfied in conformity with the economy principles.
An intriguing prospect of this approach concerns the fact that with the
theoretical status of D-structure nullified, nothing prevents Generalized

Transformation (GT) from blocking Move-a along the "least effort" guideline in

Chomsky's (1991) sense. This is because lexical insertion need not be done all
at once at a level between the lexicon and S-structure, and because GT, being
a binary operation, does not increase the "length" of a formal object, whereas

9
movement does. Other things being equal, we should prefer a recast base-
generation account to a corresponding movement one in deriving certain
syntactic/semantic dependencies. (See also Fukui (1993) for a different but
extremely interesting perspective on the issue of Economy.)
In practice, the implication has little to do with Chain formation, since a

(non-trivial) Chain is created by the singulary operation Move-a. Nor does it

hold for the concept CHAIN, because expletives introduced by there-insertion


(or do-support to a similar effect) are distinct objects in that the uniformity
condition applies only to a Chain, but not to a CHAIN (cf. Chomsky 1986a). The
situation becomes more interesting when we consider a formal object which is
by definition a pair, as in the case of operator-variable constructions. Let's take
question formation for example, and consider the foillowing two strategies of
deriving wh-dependencies:

(1) a. [x" A [x' . .wh... 1i-- [x"OP [l[x .. .wh....]

-* [x"Opil]o[x'... wh(i) ... D


b. A
[x" . .. wh... ]]- [x" whi [x' ... ti...

In (la), GT targets X', and substitutes a Q(uestion)-operator for an empty


position A external to X', which in turn binds the wh-in-situ without resorting to

Move-a. In (lb), Move-a applies instead, moving the would-be wh-in-situ into A.

(la) therefore should always have priority over (lb) on gr'ound of Economy.
On conceptual grounds, there are also a few good reasons why this should
be the case. First, the computational system of our linguistic faculty must have

10
something to operate upon. It is thus still within the conceptual necessity to
employ binary substitution. In addition, since binary substitution is the only way
to weave phrase markers into one single piece, and hence the minimal
requirement for a legitimate PF representation, it should be preferred over its
singulary counterpart. Therefore, the "courtesy" in considering lexical insertion
free is really a built-in part of Economy. If, for a priori reasons, we are not to
minimize the linguistic design into "saying nothing", "moving nothing" is
certainly the minimalist goal to achieve. We may then formulate the intuitive
idea in the following terms:

(2) Lexical Courtesy Hypothesis (LCH):

If a language may introduce an operator by binary substitution (i.e.,


Generalized Transformation), it will not resort to singulary substitution

(i.e., Move-a).

The major purpose of this chapter is to explore the consequences of the


above hypothesis, and its relation to the theory of A'-dependencies in general.
We would like to sketch three preliminary proposals with a view to
characterizing the syntactic/semantic properties of operator-variable pairs.
In section 2, we will propose that binary substitution has intrinsic priority
over singulary substitution, as required by the LCH (2). We will demonstrate
from a cross-linguistic point of view that there is an optimal design of the
architecture of wh-dependencies, which works in very much the same way as
(la) does.

11
In section 3, we will argue that only categories which introduce variables
in situ are subject to unselective binding. This move leads to a noun-adverb
distinction among wh-phrases with respect to their extraction behavior in LF.
In section 4, we will make the claim that op.,rator features are not
universally strong: Procrastination of wh-mrovement is independently motivated
by the noun-adverb asymmetry in construing Chinese wh's-in-situ, and by LF
Subjacency/CED effects displayed by Hindi wh's-in-situ.

12
1. Toward a Minimalist Design of Wh-Dependencies

A troublesome aspect of the minimalist approach concerns the fact that it is


conceptually undesirable to associate independent properties with S-Structure,
which proves to be a significant departure from the EST (extended standard
theory) model. The level of S-Structure, a derivative concept in essence, is
understood as a SPELL-OUT point of a particular derivation to PF.
Consequently, nothing should hinge upon the notion of S-Structure. Nor, it
seems, can we take locality conditions such as Subjacency to be a matter of
overt Syntax (i.e., the mapping from D-Structure to S-Structure in the sense of
EST), as originally proposed by Huang (1982), because this amounts to saying
that Subjacency ceases to be relevant after the level of S-Structure.1
The empirical motivation for Huang's proposal is well known, as illustrated
by the following argument-adjunct asymmetry of Chinese wh's-in-situ:

(3) Akiu kan-bu-qi [DP [CP Op [IP ei zuo shenme]] de reni ]?


Akiu look-not-up do what PNM person
What is the thing/job x such that Akiu despises [people [who do x]]?

(4) * Akiu xihuan [DP [CP OPi [IP Luxun weishenme xie ei]] de shui]?
Akiu like Luxun why write PNM book
What is the reason x such that Akiu likes [books [that Luxun wrote for x]]?

1 Here our reasoning may have been stretched a little bit too far. As noted by
Noam Chomsky (p.c.), making reference to the point of SPELL-OUT (i.e., S-
structure) does not necessarily associate any property with it. Consequently, the
real problem may still lie in the arbitrary ordering between Subjacency and the
level of S-structure.
13
Both (3) and (4) involve wh's-in-situ embedded in complex-NP islands. The
differences are that (3) contains a wh-argument, i.e., shenme 'what', whereas
(4) contains a wh-adjunct weishenme'why', and more importantly, that question
formation is blocked in (4), but not in (3). As Huang (1982) points out, (4)can be
ruled out along with its English counterpart (5), given that weishenme
undergoes abstract movement to the matrix Comp, and creates the offending LF
representation (6):

(5)* Whyj does John like [DP booksi [cp whichi [iP Bill wrote ti tj ]]]?

(6) * weishenmej Akiu xihuan [Dp [Cp Opi [IP Luxun t


i xie ei]] de shui]?
why Akiu like Luxun write PNM book

The deviance of (2), therefore, is identified with the kind of locality effects
typically associated with the ECP and Subjacency. The beauty of this analysis
lies in the parallelism that it envisions among languages: Sooner or later, a wh-

phrase must be related to a [+wh] Comp by Move-a. As a result, Chinese and

English wh-questions look exactly alike at LF.


Nevertheless, there remains a non-trivial problem to tackle. Let's consider
the following LF representation of (3) given the abstract movement hypothesis:

(7) ShenmekAkiu kan-bu-qi [DP[CP Opi [IP ei Zuo t]] de reni]?


what Akiu look-not-up do PNM person

The object trace of shenme 'what' is lexically governed by the embedded verb
zuo 'do', thus satisfying the disjunctive version of the ECP (Chomsky 1981,

14
Huang 1982, Lasnik & Saito 1984, 1992). On the other hand, both DP and CP
count as barriers in Chomsky's (1986b) sense, since the relative clause (i.e.,
the CP node) is not L-marked, and the DP node in turn inherits barrierhood from
the CP node. Although the Chain (shenme, t ) does not violate the ECP in its
disjunctive formulation, the Chain-formation involved undoubtedly violates
Subjacency, with both DP and CP crossed in dne link. However, no deviance is
detected in (3). For this very reason, Huang concludes that Subjacency holds in
(overt) Syntax but not in LF.
In the following sections, we would like to offer an alternative to
accommodate the asymmetry between (3) and (4) without compromising the
global status of Subjacency, which may well be an instantiation of the Economy
of derivation in the sense that crossing more than one barrier in forming one
Chain-link is considered an intolerable cost. 2 On the one hand, we will
characterize the long-distance construal in (3) as an instance of unselective
binding (cf. Heim 1982, Pesetsky 1987, Nishigauchi 1986, 1990), and hence a

2 Here the term "barrier" is used in the traditional sense, i.e., as "bounding
node". According to Chomsky's (1986b) formulation of Subjacency, even one
barrier suffices to induce marginality, since the two members of a chain link
have to be 0-subjacent (namely, without any barrier inbetween). The point can
be illustrated by the contrast between (ia,b):
(i) a. ??Whati did you remember [cP wherek she bought ti tk ]?
b. *?Whati.did you remember [DP the girl [cP who bought ti ]]?
One barrier (i.e., CP) is crossed in (ia), and hence the marginality. In contrast,
two barriers (i.e., DP and CP) are crossed in (ib), and hence the
ungrammaticality. This formulation certainly fits into the general scheme of the
minimalist approach, that is, to "minimize chain links" (Chomsky & Lasnik 1991).
Nevertheless, as Howard Lasnik (p.c.) points out, there is still substantial work
to be done before we can implement the intuitive idea presented above,
particularly before we can resolve the tension between Subjacency and
Relativized Minimality.
15
"non-Chain" (see also Li (1992), Aoun & Li (1993a,b), and to some extent,
Cheng (1991) for proposals in the same vein). On the other hand, we would
also like to maintain the insight behind Huang's treatment of (4), i.e.,
characterizing its deviance as the kind of locality effect associated with improper
Chain formation.

16
2. Singularyvs. Binary Substitution
2.1. Disintegrate Lexical Integrity

Our first proposal concerns the two fundamental operations of shaping


phrase structures under the minimalist approach; namely, binary substitution
(8a) has intrinsic priority over singulary substitution (8b):

(8) a. [x" A [x' ... wh... ] --- [x" OpoQ] [x .. wh... J


b. [x" A [x' ... wh...E]] - [x" whi [x'I*ti... .. ]

On the ground of Economy, UG should always prefer (Sa) to (8b), since


movement is employed only as the last resort, and since GT does not increase
the length of the wh-in-situ in question. (8a) thus represents the optimal dnsign
of wh-dependencies, which is unlikely to vary across languages. It also follows
from the LCH (2) that if we ever find (8b) in operation, then (8a) must have
already been implemented in some obscure way. If this reasoning proves to be
a sound one, then the key question to ask is probably not why the Chinese
sentence (3) does not display Subjacency effects, but why its English
counterpart (9) is never allowed:

(9) * Akiu despises people who do what?

If we take the question literally, potential answers are abundant: Aoun


(1986) suggests that Comp-indexing appiies at S-structure in English, and at LF
in Chinese. Since the matrix r+wh] Comp is not filled by what at S-structure, (9)
is ruled out in not complying with the selectional restriction. Similar solutions

17
can be sketched easily in terms of Spec-head agreement in the IP-CP system
(vs. the S-S' system adopted by Aoun). For instance, we may postulate to the
same effect that Wh-Criterion (10), as formulated in Rizzi (1991) (see also
Pesetsky 1982, May 1985), applies at S-structure in English:

(10) a. A wh-operator must be in a Spec-head configuration with an XO,+wh].


b. An XO[+wh] must be in a Spec-head configuration with a wh-operator.

Chomsky (1986b:52), on the other hand, maintains that selectional properties


are satisfied universally at LF, by appealing to the positive setting of the
parameter (11) in English, based on the assumption that vacuous movement is
optional at S-structure:

(11) At LF, wh-phrases move nonvacuously only to a position occupied by wh-.

Under (11), what is not allowed to undergo (nonvacuous) LF movement in (9),


since the matrix CP Spec is not occupied by another wh-phrase. As a result, the
sentence is correctly ruled out. It is also possible to sketch a minimalist solution
without resorting to the notion of S-structure: Given that operator features such
as [+wh] are strong in English, procrastination of wh-movement is not allowed.
Consequently, the derivation of (9) crashes at PF, because the [+wh] feature on
what is visible but unchecked.
Our question, however, has a quite different connotation in the face of the
LCH, which amounts to asking why English never takes advantage of the
design represented by (8a). The answer, in our opinion, is that English does
implement the design, but in a miniature scale. To see this, let's compare the

18
following paradigms:

(12) a. wh-words b. pronominals


wh-o wh-en th-ey th-en
wh-om wh-ere th-em th-ere
wh-at th-at

By comparing (12a) with (12b), it is not difficult to see that English wh-words and
pronominals are more or less built on the same materials except that the prefix
for pronominals is th- instead of wh-.3 Nevertheless, there is a crucial distinction
between these two morphemes: Th-, for obvious reason, should be regarded as
a reduced form of English definite article the, capable of licensing the indefinite
morphemes it attaches to (i.e., ey, en, em, ere, and at). Wh-, on the other hand,
does not seem to act as a determinant of quantificational force, as evidenced by
the free relative construals of wh-words:4

3 Similar intuitions have been pursued as early as Chomksy (1964), Katz &
Postal (1964), and Klima (1964), where who and what are analyzed as
[WH+someone] and [WH+something] respectively. See also Kuroda (1965).
4 As Howard Lasnik (p.c.) observes, free relative constructions like (i) do not
fall under the category discussed here:
(i) I'll eat what you cook.
The usage of the free relative what in (i) is considered to be definite by C. L.
Baker (1989), in contrast to the indefinite usage exemplified by (13a). The
distinction can be illustrated by the following paraphrase of (i):
(ii) I'll eat the food that you cook.
It follows that there may well be a (non-overt) definite counterpart of -ever
associated with the relative wh in question As noted again by Howard Lasnik,
our treatment here bears close resemblance to that of Chomsky (1975:434),
where who is taken to be [WH+(s)he], what [WH+it], and so forth. Consequently,
there seem to be empirical bases for both the [WH+pronominals] and
[WH+indefinites] analyses. If our proposal presented below is correct (cf. (14a)
and (19)), the difference between the definite and indefinite usages then lies in
the different operators involved in the internal structures of wh-words.
19
(13) a. free relative wh's b. pronominals
wh-o-ever wh-en-ever * th-ey-ever * th-en-ever
wh-om-ever wh-er(e)-ever * th-em-ever * th-er(e)-ever
wh-at-ever * th-at-ever

(13a) shows that wh- does not block binding from the suffix -ever, which
contributes universal force to the indefinites, as illustrated by (14a). As a result,
whoever can be paraphrased directly as 'anyone', whatever as 'anything',
whenever as 'anytime', and so forth. In contrast, pronominals cannot be suffixed
by -ever, as shown by (13b). This indicates that th- blocks the binding construal
between -ever and the indefinite in (14b), just as its determiner counterpart
might do in a full DP:

(14) a. NO b. *NO

NO -everx NO -everx

wh- ind.(x) th- ind.(x)

A similar pattern can also be found in a class of adverbials built on wh-words:

(15) a. wh-adverbials b. pronominals


some-wh-at * some-th-at
some-wh-ere * some-th-ere

Here the binding relation holds between the existential operator some and the
indefinite morphemes (i.e., at and ere), as shown by (15a). As it turns out,
somewhere can be read as 'in some place' somewhat 'to some extent' (though

20
it can also be construed in parallel with whatever, meaning 'something', as in
He is somewhat of an artist ). The same construal is not available to the
corresponding pronominals in (15b). The cause of this asymmetry again seems
to iie in the (strong) quantifier status of th-, which render -ever and some-
vacuous quantifiers, as illustrated by (14b) and (16b) respectively:

(16) a. AdvO b. * AdvO

some-x AdvO some-x AdvO

wh- ind(x) th- ind(x)

The contrast between (13a,b), as well as that between (16a,b) is therefore


reminiscent of Reinhart's (1992) observation that DPs headed by which may be
treated as a function variable subject to unselective binding, which option is
never available for definite DPs. We will return to and elaborate on this point
later.
With the prospect created by (14a) in mind, we may put the design (8a) into
work. A natural suggestion here is that interrogative wh's have a Q(uestion)-
operator instead of -ever as the relevant binder, as illustrated by (17):5

5 Here we make no commitment as to which way the Q-operator should


adjoin to the wh-word in question. For the ease of comparison, we put it to the
right of No, although this should not be possible if Kayne (1993) is right about
the Linear Correspondence Axiom (LCA). Since there is always some
arbitrariness about the subword-level syntax, as we have already seen in free
relatives such as what-ever (vs. some-what), we will leave the issue open here.
21
(17) NO

NO Opx IQ]

wh- ind.(x)

This move can be achieved by employing binary substitution, which targets the
lower NO, and extends the target by inserting a Q-operator as its specifier. We
then have a ready answer for why (9) is not a possible English sentence. As
illustrated by (18a), since what in itself is an operator-variable pair (recall that
(8a) has been implemented below the XO-level, as in (17)), unary substitution is
uncalled-for.6 Even if it does apply, as it is costless, the Q-operator dangling
alone in the CP Spec results in vacuous quantification:

(18) a. [CPA [IP ... what-Op[al ... ]] -- x- [cP Op[al [IP ... what-Oplo]...]]

b. [CP A [IP ... what-OpQll ... ] - [cp [what-Op[Qol]i iP ...ti...Df

The rest of the story is essentially the same as given above: Since the [wh]
feature is strong in English, singulary substitution must apply before SPELL-
OUT to make sure that what is in the matrix CP Spec for feature-checking, as
illustrated by (18b). (9) is thus ruled out in failing to check the strong feature of
what in overt syntax.

6 Another possibility, as entertained by Noam Chomsky (p.c.) is assume that


English does not allow Q-operators at all. This proposal, though stipulative in
nature, is not further complicated by the tension between morphology and
syntax. For a similar view, see Tsai (1992a), where the presence of Q-operators
in Chinese in correlated to the positive setting of the null topic parameter in
Huang's (1984) sense.
22
As for relative wh-words, we may either leave the structure as it is, in which

case relative wh's are essentially null operators plus a set of 4-features, or insert

a null operator in the specifier position, serving as a medium for identification


(cf. Chomsky 1986a), as illustrated below:

(19) No

NO Opx

wh- ind.(x)

Either way we make the right prediction that relative wh's are not quantifiers.
Rather, they take scope positions for defining the domain of (syntactic)
predicates. It is also in this sense that "free relatives" is an ironic term for wh-
words like whoever, since they are really "bound" relatives, as much as ordinary
relative wh's are free.
To sum up, if our position proves to be defendable, then the feature [wh]
should be conceived as a purely morphological device, whose function is to
create configurations for both predication and question-formation through
feature-checking on the Spec of CP.

2.2. Dou, dou, dou, and dou

As a suggestive comparison, we would like to point out that Chinese dou


'all' has a few different types of usages, each of which roughly corresponds to
one of the following English words: all, always, already, and also:

23
(20) a. tamen /*ta dou zou-le.
they / (s)he all leave-Prf
They all left.
b. zheli shei dou bu xihuan Akiu.
here who all not like Akiu
Here nobody likes Akiu.
(21) Akiu dou qu nali nao geming?
Akiu all go where incite revolution
Where does Akiu go to incite revolution all (the time)?
(22) (the-ci geming), Akiu dou qu-le na-,,.
This-CL revolution Akiu all go-Prf where
Where has Akiu already been to (in this revolution)?

(23) Akiu (lian) yi-fen qian dou / ye yao.


Akiu even one-penny money all/also want
Akiu also wants (even) one penny.
The most common usage of dou is acting as a universal operator quantifying
over plural NPs or wh-NPs to its left, as shown by (20a,b) respectively. 7 But

7 An apparent counterexample to the above observation comes from cases


where NPs under quantification take singular determiners such as (yi-)zheng-
ge'(one)-whole-CL', as evidenced below:
(i) a. (yi-)zheng-ge fangzi dou shao-diao-le.
one-whole-CL house all burn-down-Prf
The whole house has burnt down.
b. Akiu ba (yi-)zheng-ge fangzi dou shao-diao-le.
Akiu BA one-whole-CL house all burn-down-Prf
Akiu has burnt down the whole house.
As suggested by Cheng (p.c.), the anomaly may result from some kind of
partitive construals. The definite readings reflected by the English translations
also support her conjecture. If this is indeed the case, then (ia) should be read
as 'all parts of the house have burnt down", and (ib)as 'Akiu has burnt down all
parts of the house'.
24
what is more intriguing in the context of our discussion concerns the adverbial
construals presented above. Let's start with (21) and (22). First note that there is
no plurality restriction on the subject Akiu. It is thus unlikely for dou to quantify
over individuals. Rather, the aspectual interpretations indicate that dou actually
induces universal quantification over time segments (or temporal-spatial slices
of an event in terms of intensional semantics). The difference is that the reading
tends to be collective in (21), but distributive in (22), as illustrated by (24a,b)
respectively:

(24) a. For all x, x a time, where does Akiu go to incite revolution at x?


b. For each x, x a time (during the period of this revolution), where has
Akiu been to at x?

This move captures the fact the answer to (21) has to be a unique place, while
that to (22) is expected to be a list of locations, since the speaker is asking for a
choice for each set of propositions associated with one distinct time segment. 8

See also Lee (1986), Chiu (1990), and Cheng (1991,1993) for discussions
and debates over the origin of dou and its S-structure position.
8 As Lisa Cheng (p.c.) points out, there is still a problem concerning why dou
is ambiguous in regard to quantification over time segments. One way to look at
it is to appeal to scope interaction and the aspectual distinction between (21)
and (22). That is, we may assume that the generic tense tends to trigger the
wide-scope construal of dou over nali 'where' in (21), while the perfective
aspect of (22) suppresses the same reading in favor of the narrow-scope
construal. We may also take the ambiguity to be a lexical one. Under this view,
there is no scope interaction even in the following example (cf. May 1985):
(i) What did everyone buy for Max?
a. What is x, x a thing, such that for every y, y a person, y bought x for
Max?
b. For every y, y a person, what is x, x a thing, such that y bought x
for Max?
Everyone in (i) is simply ambiguous between collective and distributive, i.e.,
25
(23), on the other hand, is a totally different story. As shown above, dou
alternat3s with ye 'also', often in conjunction with an optional focusing adverb
lian 'even'. Here dou certainly does not quantify over the object yi-fen quian
'one penny'. Namely, (23) does not mean 'Akiu wants even every penny'.
Rather, there seems to be a universal quantification over the contrast (pre-
established) set implicated by the semantics of lian 'even' (see also Rooth
1985). As noted by Jim Huang (p.c.), the smallness of the amount of money
(merely one penny) is emphasized by contrasting it with all the larger amounts
of money which Akiu would also love to have. According to Horn (1988), the
English counterpart of (23) has the following semantics.

(25) a. John wants even one penny.


b. Even (x=the amount of one penny, want (John,x))
Presupposition: (3 y) (yx & want (John,y))
Assertion: want (John,x)

We may then treat dou and ye as the triggers of universal quantification over the
contrast set associated with the presupposition of lian. As a matter of fact, as
Alec Marantz (p.c.) points out, the practical interpretation of (25a) comes very
close to that of its Chinese counterpart, as illustrated by (26b):

between the all-type reading and the each-type reading (also cf. Lasnik & Saito
1992). The same observation applies to dou in (21) and (22).
26
(26) a. Akiu lian yi-fen qian dou / ye yao.
Akiu even one-penny money all / also want
b. Lian (x=the amount of yi-fen, yao (Akiu,x))
Presupposition: (Vy) (yox & yao (Akiu,y))
Assertion: yao (Akiu,x)

Namely, one penny is contrasted with ALL the larger amounts of money, rather
than just SOME amount of money. If the above semantics proves to hold across
languages, then dou and ye may well be the syntactic reflexes, rather than the
actual triggers, of the universal quantification at issue here.
In the light of the above observation, English adverbs such as always,
already, and also may well be analyzed as miniature operator-variable pairs, in
that they are all prefixed by al-, a reduced form of all.9 Although this line of
speculation still calls for a fine-grained lexical semantics, the pattern is already
there. That is, while English implements the design (8a) word-internally,
Chinese does it at the sentence level.

23. Interrogative construals in Chinese bare conditionals

if our analysis is on the right track, Chinese wh-questions may well


embody the optimal design in its grandest scale. That is, (3) has (27) instead of

9 As noted by Ken Hale (p.c.), Dutch al is also ambiguous between 'all'


and 'already'. In addition, it can form compounds with nominal stems, e.g., al-tijd
'anytime', al-ler-zijds 'anywhere', al-ler-wegen 'always'.
27
(7) as its LF:10

(27) [cP Opx [q] [i Akiu kan-bu-qi [DP [CP Opi [IP ei zuo shenme(x) ]] de reni ]?
Akiu look-not-up do what PNM person
What is the thing/job x such that Akiu despises 1people [who do x]]?

Since the operator-variable pair (Opjo], shenme) is built by binary substitution

and unselectively binding, and since Move-a is not involved, naturally we do

not expect any Subjacency effect. As a result, nothing hinges upon the notion of
S-structure in this solution.
Probably the most solid showing of this sentence level of binding comes
from Chinese bare conditionals. As noted by Cheng & Huang (1993), a pair (or
pairs) of wh's-in-situ separately located in two clauses induces a conditional
construal without any lexical marker like ruguo 'if' or yaoshi 'in case that':

(28) a. shei lai, shei chi.


who come who eat
If x comes, x eats (it).

b. shei xian lai, shei (jiu) xian chi.


who first come who then first eat
If x comes first, x eats first.

Since this type of conditional can be so "bare" as (28a,b), it is not difficult to see
that there must be an abstract (necessity) operator which binds both the wh's-in-

lo For a non-movement analysis of the long-distance construals of wkh-


arguments and the postulation of an abstract Q-operator in sentential
projections, see Li (1992), Aoun & Li (1993a,b), Tsai (1992c, 1993), and to a
considerable extent, Cheng (1991).
28
situ and enables the construal. The resulting logical representations thus have
a classical donkey outlook (cf. Helm 1982):

(29) a. Vx (x comes -- x eats it)


b. Vx (x comes first-* x eats first)

They further point out that the second wh-in-situ in the consequent clause
cannot be treated as an E-type pronoun in Evans' (1980) sense, since it is in
complementary distribution with typical pronominals and demonstratives, as
evidenced by the contrast between the bare conditional (30a) and theruguo-
conditional (30b):

(30) a. shei xian lai-le, shei/*na-ge-ren/ta/pro jiu yao qing-ke.


who first come-Inc who/that-CL-preson/(s)he then must invite-guest
If x happens to come first, then x must play the host.
b. ruguo shei xian lai-le, na-ge-ren/ta/pro/*shei jiu yao qing-ke.
if who first come-Inc that-CL-person/(s)he/who then must invite-guest

As a result, (28) is most likely to be an instance of unselective binding, just as


Heim (1982) originally proposes for donkey sentences. Their position is
supported by the fact that the pairing not only can be multiple, but also can be
interchangeable, as long as the wh's-in-situ involved are indentical:

(31) shei qin-le shei, sheijiu yao qu shei.


who kiss-Prf who who then must marry (male to female) who
a. If x has kissed y, then x (male) must marry y (female).
b. If x has kissed y, then y (male) must marry x (female).

29
This is fully expected since the binders in question are claimed to be
unselective. We may thus assign the following two logical representations to
(31a,b) respectively:

(32) a. Vx Vy (x has kissed y -- x must marry y)


b. Vx Vy (x has kissed y - y must marry x)

By incoporating the notion of tripartite structure (Kemp 1981, Heim 1982,


Chierchia 1992) into the Parallelism Constrain on Operator Binding (PCOB) in
Safir's (1985) sense,"1 they propose the following princ ' further to capture
the identity condition on the pair(s) of wh's-in-situ:

(33) Revised PCOB:


In a tripartite structure of quantification, Q [A] [B], [xi, x2 , ... xn], where
n>l, are variables in A. For every variable in A, there must be an identical
variable in B.

For the purpose of this chapter, let's twist the bare conditionals a little bit,
i.e., making the numbers of wh's-in-situ unevzol in the antecedent and
consequent clauses. The result is most curious. Sentences such as (34) and
(35) are not ruled out, with the "stranded" wh's-in-situ construed as
interrogative:

11 The PCOB is formulated as follows:


If O is an operator and x is a variable bound by O, then for any y, y a
variable of O, x and y are [ a lexical].
30
(34) shei qin-le shei, shei jiu dian-dao-bu-yi (ne)?
who kiss-Prf who who then upside-down-no-end Qvh

a. Who is the person x such that if x has kissed y, then y will be


hopelessly overwhelmed?(The first wh stranded)
b. Who is the person y such that if x has kissed y, then x will be
hopelessly overwhelmed?(The second wh stranded)

(35) shei xian mai fanzi, shei jiu jia shei (ne)?
who first buy house who then marry (female to male) who Qi

a. Who is the person x such that if y buys a house first, then x (female)
will marry y (male)? (The second wh stranded)
b. Who is the person y such that if x buys a house first, then x (female)
will marry y (male)? (The third wh stranded)

In (34), we have two wh's-in-situ in the antecedent clause, but only one in the
conseqent clause. Either the first wh or the second wh can induce a matrix
question, with the other paired with the third wh in the consequent clause,
forming a bare conditional, as illustrated by (34a,b) respectively. The situation
with (35) is the other way around. The scenario can be set in ancient time, when
wealth is everything in marriage. There is only one wh-in-situ in the antecedent
clause, but two in the consequent clause: Either the second wh or the third wh
can induce a matrix question, with the other paired with the first wh in the
antecedent clause, as illustrated by (35a,b) respectively.
In the light of our view presented in (27), this phenomenon is hardly
surprising, since there is neither worry about Sujacency or the CED associated
with the adjunct island (i.e., the ruguo-clause), nor concern about Relativized

31
Minimality or the ECP in the matrix. The situation, on the other hand, becomes
less clear when we reconsider sentences like (28b) in the new light:

(36) a. shei xian lai, shei jiu keyi xian chi ne?
who first come who then can first eat Qm
Who is the person x such that if x comes first, then x is allowed
to eat first?
b. Akiu xiang-zhidao [shei xian lai, shei jiu keyi xian chi].
Akiu want-know who first come who then can first eat
Akiu wonders [who is the person x such that if x comes first, x is
allowed to eat first].

When a touch of "privilege" is added to the predicate xian chi 'eat first', the
interrogative reading becomes available with the conditional construal
remaining intact, as shown by (36a). This reading is most salient when the bare
conditional is embedded as an indirect question, as shown by (36b).
The solution, in our opinion, lies in a more articulated logical
representation of (28b) offered by Cheng & Huang (1993), as illustrated below:

(37) For all (x,s(ituation)) (if x comes first ins), (x eats first in s)

Here the situation or spatiotemporal variable is spelled out, and the semantics
shared by bare conditionals and ruguo-conditionals is also captured. More
importantly, this analysis allows us to solve the puzzle straightforwardly. That is,
by delegating the conditional construal to the pair of situation variables, there is
plenty of room left for the interrogative construal on the part of wh's-in-situ. As a
result, the proposed Q-operator is able to "cut in" and license the question

32
readings of (36a,b), as illustrated below:

(38) Qx [x a person] Vs [s a situation] if x comes first in s, x is allowed to eat first ins

Our observation thus lends strong support to the analysis of (37) in turn.
At this stage, it is tempting to suggest that maybe the conditional construal
has nothing to do with the wh'-in-situ, since they appear to have their own life. In
fact, this position is not totally hopeless at first glance:

(39) a. laoshi shang-ke renzhen, xuesheng *O(iu) hui yong-gong.


teacher give-lecture attentatively student then will exert-effort
If teachers teach attentatively, then students will study hard.
b. Akiu yong-gong, ta/pro*(jiu) shang-de-liao daxue.
Akiu exert-effort he then go-can-finish college
IfAkiu studies hard, then he can get into college.

(39a,b) show that without any conditional marker or wh-in-situ, conditional


construals are still possible. However, the presence of jiu 'then' becomes
obligatory, which seems to be a physical reminder of the missing conditional
marker. Similar situations also obtain in some apparent violation of the revised
PCOB, as originally observed by Yu (1965):

(40) shei yong-gong, na-ge-ren/ta/pro *(jiu) shang-de-liao daxue.


who exert-effort that-CL-person/(s)he then go-can-finish college
If anyone/someone studies hard, then (s)he can get into college.

The fact that (39a,b) are well-formed therefore does not indicate that
universal quantification over situations alone can license the conditional

33
construal in quesiton. Rather, it confirms Cheng & Huang's conjecture that
Chinese allows an abstract conditional operator. Consequently, (39e,b), as well
as (40), should be interpreted as if there is a conditional marker ruguo 'if'
standing in the sentence-initial position. Furthermore, while an indefinite wh-in-
situ may occur unpaired in an antecedent clause, as in (40), presumbly licensed
by the unseen ruguo, consequent clauses in general block this construal, as
exemplified below:

(41) Akiu yong-gong, shei jiu shang-de-liao daxue?


Akiu exert-effort who then go-can-finish college
Akiu studies hard, then anyone/someone can get into college.
a. #If
b. Who can get into college if Akiu studies hard?

As a result, (41) can only be construed as interrogative. It is then not surprising


to see that the interrogative construal in (38) does not allow a "paired" reading
(i.e., with the pair of variables referring to distinct persons). In other words,
representations like (42) are never possible for (36a,b):

(42) Qx,y [x,y a person] Vs [s a situation] if x comes first in s,y is allowed to


eat first ins

This suggests that even though the wh's-in-situ in question are subject to
independent construals, their licenser (i.e., the Q-operator) still observes the
same principle as the necessity operator (i.e., the revised PCOB (33)).
To get the whole picture, we need to further consider constructions such as
(43), where paired interrogative readings are actually possible. Here we
repalce qu 'marry (male to female)' in (31) with zou 'beat', to make a "love

34
triangle" scenario more plausible. For ease of exposition, we will omit situation
variables, and reserve the formulae "Qx,z" and "Qy,z" exclusively for paired
construals of distinct wh's from different clauses. In addition to the expected
double conditional readings (cf. (32)), (43) has the following question readings:

(43) (cai-cai-kan) shei qin-le shei, shei jiu hui zou shei (ne)?
try-to-guess who kiss-Prf who who then will beat who Qw

a. (if x has kissed y, then x will beat y)


b. (if x has kissed y, then x will beat y)
C. (if x has kissed y, then y will beat x)
d. (if x has kissed y, then y will beat x)

e. Vx Qy,z (if x has kissed y, then x will beat z)


f. Qx,z Vy (ifx has kissed y, then z will beat y)
g. Vx Qy,z (if x has kissed y, then z will beat x)
h. Qx,z Vy (if x has kissed y, then y will beat z)

i. Qx Qy (if x has kissed y, then x will beat y)


j. Qx Qy (if x has kissed y, then y will beat x)
k. Qx Qylz (if x has kissed y, then x will beat z)
I. (if x has kissed y, then z will beat y)
Qx Qy,z
m. (if x has kissed y, then z will beat x)
n. Qxz Qy (if x has kissed y, then y will beat z)

As one might expect, even with the "love triangle" scenario in mind, it is not easy
to get all of the readings in (43e-h) and (43k-n) due to ensured complexity. To
exhaust all the possibilities, one needs a carefully constructed scenario to pin
down each reading. Let's take two most natural readings for example: The
answers to (43g) and (43m) could be (44a) and (44b) respectively, where

35
Diaochanand LObu are a couple and Donzhuo is the villain:

(44) a. shei qin-le Diaochan, LUjbu jiu hui zou shei.


who kiss-Prf Diaochan L"ibu then will beat who
If x has kissed Diaochan, then Li*bu will beat x.
b. Donzhuo qin-le Diaochan, L*bu jiu hui zou Donzhuo.
Donzhuo kiss-Prf Diaochan LU3bu then will beat Donzhuo
If Donzhuo has kissed Diaochan, then LUibu will beat Donzhuo.

On any account, our main purpose here is to point out that paired question
construals (i.e., construals involving distinct wh-variables) are allowed only if at
least one pair of indentical wh-variables is licensed, either by a necessity
operator or by a Q-operator. In other words, abstract situation variables do not
have independent status in licensing bare conditionals, and the following
representations are blocked along with (42):

(45) a. Qx,z Qy,k (if x has kissed y, then k will beat z)


b. Qx,z Qy,k (if x has kissed y, then z will beat k)
c. Qx,z Qy,k (if y has kissed x, then k will beat z)
d. Qx,z Qy,k (if y has kissed x, then z will beat k)

The implication is far-reaching. As we have seen in (34) and (35), the


question construal does not observe the revised PCOB when the conditional
construal is not at stake.12 The same observation applies to (36a,b) and (43).

12 One might notice that the readings given in (34) and (35) do not exhaust all
the logical possibilities. For some reason, it is difficult to get the following two
readings from (34) and (35) respectively:
(i) a. Who are the persons x,y such that if x has kissed y, then y will be
hopelessly overwhelmed?
36
This indicates that there is no "archetype" behavior for unselective binding:
Everything is conditioned by logical necessity imposed by principles such as
the revised PCOB, which happens to apply on the sentence level due to the
morpho-syntactic makeup of Chinese. In other words, unselective binding is just
binding, definable in terms of the notion of c-command. By reflecting upon the
specficity effects manifested by lexical binding in English (cf. the contrasts
between (13a,b)), we find similar traits. That is, lexical binding is just binding,
subject to general principles such as the specificity condition (Higginbotham &
Fiengo 1981), and definable in an optimal and minimal way. We will allude to
this point later when we are prepared to sort out some potential problems with
the LCH.

2.4. Japanese mo-quantification and Chinese dou-quantification

Along the line sketched above, we would expect to find languages in


between, where the Q-operator is located in the vicinity of regular determiners,
say, the Spec of DP. This possibility has actually been entertained by Watanabe
(1991), based on the variety of indefinite contruals of Japanese wh's-in-situ.
First consider the following paradigms:

b. Who are the persons x,y such that if x has kissed y, then x will be
hopelessly overwhelmed?
Our best guess is that the readings require to pair two uneven dependencies,
i.e., one with a tripartite structure and the other without, resulting in the
deviance.
37
(46) a. interrogative construals b. universal construals
dare 'who' dare-mo 'everyone'
nani 'what' ?nani-mo 'everything'
doko 'where' doko-mo 'everywhere'
itsu 'when' itsu-mo 'whenever'
naze 'why'

c. polarity construals d. existential construals


dare-mo 'anyone' dare-ka 'someone'
nani-mo 'anything' nani-ka 'something'
doko-mo 'anywhere' doko-ka 'somewhere'
itsu-ka 'sometime'
naze-ka 'for sc - reason'

Built upon Kuroda's (1965) observation that Japanese wh-words (indeterminate


pronominals in his terms) behave rather like unbound variables, Nishigauchi
(1986,1990) makes the proposal that morphemes such as -mo and -ka should
be analyzed as unselective binders in the sense of Lewis (1975) and Heim
(1982). He further demonstrates that these morphemes behave rather like
prepositions or determiners than part of a word. For instance, ka may switch its
word order with prepositions like kara 'from', inducing subtle semantic
distinction:

(47) a. dare-ka-kara henna tegami-ga todoi-ta.


who-some-from strange letter-Nom arrived
A strange letter came from somebody.
b. dare-kara-ka henna tegami-ga todoi-ta.
who-from-some strange letter-Nom arrived
A strange letter came from god knows who.

38
Mo, on the other hand, can be further attached to adjunct clauses and complex
NPs:

(48) a. [dare-ga ki-te]-mo, boku-wa aw-a-nai.


who-Nom come all I-Top meet-not
For alix, if x comes, I will not meet (x).
b. [[dare-ga kai-ta] tegami]-ni-mo onazi kota-ga kai-te-at-ta
who-Nom wrote letter in all same thing-Nom written-was
For all x,y, x a person, y a letter x wrote, the same thing was written in y.

The conditional construals of (48a,b) are reminiscent of a variety of bare


conditionals in Chinese, which patterns with ruguo-conditionals in disallowing a
wh-in-situ in the consequent clause (cf. Cheng & Huang 1993):

(49) shei xian lai, na-ge-ren/ta/pro/*shei dou yao qing-ke.


who first come that-CL-preson/(s)he/who all must invite-guest
Ifx comes first, x must play the host.

As shown above, the only difference between the dou-conditional (49) and a
genuine bare conditional is that the optional connective jiu'then' is replaced by
dou 'all'. Their semantics nonetheless differ drastically: Only demonstratives
and E-type pronominals, but not wh's-in-situ, are allowed in the consequent
clause of (49). Moreover, the other two usages of mo also remind us of the
peculiar behavior of dou mentioned above (data from Nishigauchi 1990):

39
(50) a. John mo ki-ta.
John also came
(In addition to other persons,) John also came.
b. John-ga ki-te mo, ...
John-Nom come even-if
Even if John comes, ...

As exemplified below, dou has an obscure usage hidden behind the usual
universal interpretation:

(51) ni-men dou lai ba!


you (pl.) all come Exc
a. Come! All of you!
b. Come along! you guys!

(51a) represents a typical reading of dou, quantifying over plurals. It is the


second reading (51b) which bears resemblance to (50a). The difference is that
dou requires a plural subject in both readings. When a singular subject is
substituted, only ye 'also' may appear:

(52) ni ye lai ba!


you also come Exc
You come along!

(Lian) .. . dou 'even ... all' constructions such as (23) also allow construals
parallel to (50b), where the focus position is occupied by a CP instead of a DP:

40
(53) Dongzhuo (a), (lian) [cP Labu lai] dou/ye tai-bu-dong.
Dongzhuo Top even Lebu come all/also lift-not-move
Even if LObu comes, (he) cannot lift Dongzhuo.

Here universal Quantification again applies over the contrast set of the
proposition LObu lai, resulting in the focus construal. (53) thus can be
paraphrased as 'Even LObu cannot lift Dongzhuo, let alone all the others'. The
same analysis seems to hold for its Japanese counterpart (50b) as well.
This parallelism seems to break down when we consider the contrast
between (55a,b): As Nishigauchi observes, corresponding to the distinction
between every and free-choice any in (54a,b) (cf. Hornstein 1984), there is an
asymmetry between mo- and demo-quantification with respect to their ability to
license pronominals across sentence boundaries:

(54) a. Take every number. *1will divide itj by three.


b. Take any numberi. I will divide iti by three.

(55) a. Dono sakanai-mo mot-te ki-te kudasai.


which fish all carry come please
*Sorei-o ryoori-si-te agemasu kara.
it-Acc cook-for-you because
Bring in every fishi :Iwill cook iti for you.

b. Dono sakanal-demo mot-te ki-te kudasai.


which fish even carry come please
Sorei-o ryoori-si-te agemasu kara.
it-Acc cook-for-you because
Bring in any fishi : I will cook iti for you.

41
The contrast in question thus bear some resemblance to the asymmetry
between bare conditionals and duo-coditionals in Chinese.
On the other hand, although mo-conditionals appear to pattern with bare
conditionals in (55b), mo does not license an extra wh-in-situ in the main
clause, as shown by (56) (cf. (48a)), any more than its Chinese counterpart dou
does in (57) (cf. (49)):

(56) *[darei-ga ki-te]-mo, boku-wa darei-ni aw-a-nai.


who-Nom come all I-Top who-Dat meet-not
For all x, if x comes, I will not meet x.

(57) *[sheii lai], wo dou bu jian sheii.


who come I all not meet who
For all x, if x comes, I will not meet x.

Therefore, mo and dou behave similarly at least in two respects: First, they may
license a wh-in-situ from a detached position. Second, they only take scope
over the antecedent clause in a conditional construction. In other words, they
never license a wh-in-situ in a consequent clause. This point can alsc be
illustrated by comparing the following two constructions:

42
(58) a. na-ge renj/k /taj/*k /sheij/,k [CP sheik lai] dou hui lian-hong.
that-CL person/(s)he/who who come all will face-red
Whoeveri comes to visit, that personi /(s)hei will flush.
b. [cp sheik lai]i, na-ge renj/k /taj/k /sheij!k tidou hui lian-hong.
who come that-CL person/(s)he/who all will face-red
Whoeveri comes to visit, that personi /(s)hei will flush.

First we leave out null subjects, since there is no way to tell their positions
relative to the focused constituent in question. As shown by (58a), definite
subjects such as na-ge ran 'that person' and ta '(s)he' cannot be coreferential
with the wh-in-situ in the adjunct (antecedent) clause, presumably due to
Principle C violations. In contrast, when the antecedent clause is preposed to
the sentence-initial position, as in (58b), referential contruals become available.
Unselective binding construals, on the other hand, are blocked in both cases, 13
which suggests that dou never widens its scope by undergoing LF QR to license
a genuine bare conditional. In other words, (58a,b) can never have the
following donkey-type representation:

(59) doux (x comes -- x will flush)

It is also instructive to note that this "scope rigidity" follows straightforwardly from
the Syntax-LF isomorphism in Huang's (1982) sense.
Besides, as pointed out by Hiro Ura and Masa Koizumi (p.c.), (55b)

13 The j-indexed readings of matrix shei in (58a,b) result from marginal


interrogative construals, as illustrated below:
(i) ?Who is the person x such that whoever comes to visit, x will flush?
43
improves when we replace the singular pronoun sore with its plural counterpart
or an empty pronominal. Consequently, the reading of mo might be collective
rather distributive in this particular case, which in turn suggests that the
similarity between (55b) and Chinese bare conditiona! i- only apparent.
The real difference, it seems, still lies in their structural properties. Ka and
mo behave like determiners or prepositions (cf. Nishigauchi 1986,1990): They
cannot be stranded by scrambling the constituents over which they take scope,
as evidenced by the contrast between (60b,c), as well as that between (61 b,c):

(60) a. Dare-mo-ga nani-ka-o tabe-te-iru.


everyone-Nom something-Acc eating-be
Everyone is eating something.
b. [Nani-ka-o]i dare-mo-ga t tabe-te-iru.
something-Acc everyone-Nom eating-be
c. *Nanii dare-mo-ga ti-ka-o tabe-te-iru.
what everyone-Nom some-Acc eating-be

(61) a. Dare-ka-ga dare-mo-o aisi-te-iru.


someone-Nom everyone-Acc love-be
b. [Dare-mo-o]i dare-ka-ga ti aisi-te-iru.
everyone-Acc someone-Nom love-be
c. *Darei dare-mo-ga ti -mo-o aisi-te-iru.
who someone-Nom every-Ace love-be

In contrast, dou can be and in fact must be stranded in the presence of focus
movement, as shown by the contrast between (62b,c), as well as that between
(63b,c):

44
(62) a. Akiu shei dou xiangxin.
Akiu who all trust
Akiu trusts everyone.
b. sheii, Akiu ti dou xiangxin.
who Akiu all trust
c. *[shei dou]i, Akiu t xiangxin.
who all Akiu trust

(63) a. Akiu [cp shei lai] dou hui lian-hong.


Akiu who come all will face-red
Whoever comes to visit, Akiu will flush.
b. [cpshei lai]i, Akiu t1 dou hui lian-hong.
who come Akiu all will face-red
c. *[[cpshei lai] dou]i, Akiu ti hui lian-hong.
who come all Akiu will face-red

As shown by (62a) and (63a), dou usually takes scope over a constituent
immediately to its left. Apparent exceptions to the above generalization such as
(62b) and (63b) thus can be attributed to local focus movement. (62c) and (63c)
indicates that dou may behave like a clitic, but it is not a morphological suffix or
a determiner.
A side comment here concerns an interesting comparison between dou-
constructions and (lian) ... dou constructions. The first noticable thing is that the
presence of lian 'even' becomes obligatory when the focus position is occupied
by a wh-in-situ, as in (64a), or by a CP containing a wh-in-situ, as in (64b):

45
(64) a. Akiu *(lian) shei doutye xiangxin (ne)?
Akiu even who all/also trust QM
Who is the person x such that Akiu trusts even x?
b. Akiu *(lian) [cp shei lai] dou/ye hui lian-hong (ne)?
Akiu even who come all/also will face-red QC
Who is the person x such that Akiu will flush even if x comes?

The cause could be functional, since (64a,b) can only be construed as


interrogative, in contrast to (62) and (63). The obligatory presence of lian thus
disambiguates the potential confusion. The exclusive it :ogative construal
also lends support to our view that dou actually quantifies over the contrast set
of the focused constituent in (lian) ... dou constructions (cf. section 2.2), not the
constituent itself. Otherwise, the readings of (64a,b) should be universal, just
like (62) and (63).
The conclusion thus appears to be that while Japanese (de)mo-
quantification patterns with Chinese dou-quantification in almost every aspect of
its semantics, they differ in morph-syntactic terms: (De)mo behaves as a part of
the constituent over which it takes scope, whereas dou appears to a sentential
adverbial cliticized leftward to whatL ier it can quantify over (see also Lee 1986,
Cheng 1991,1993).
The asymmetry becomes even clearer when we consider the fact that
Chinese-style bare conditionals are nowhere to be found in Japanese, as
evidenced by (65):

46
(65) * darei-ga ki-te, boku-wa darei-ni aw-a-nai.
who-Nom come I-Top who-Dat meet-not
For all x, if x comes, I will not meet x.

This suggests that Japanese does not allow an abstract necessity operator on
the sentence level, which in turn substantiates our conjecture that Japanese
stands in between Chinese and English in tems of the maneuverability of
(unselective) binding. Given the equation that the degree of binding
maneuverability refects the structural height of binders, we may well
hypothesize that Japanese Q-operators locate in the Spec of PP or DP, in the
vein of Watanabe (1991), Tsai (1992b), and Aoun & Li (1993b).

47
2.5. A conceptual problem and an empirical solution

A conceptual problem with the Lexical Courtesy Hypothesis (2), as raised


by Noam Chomksy (p.c.) is that although the binary substitution strategy has
initial advantage over its singulary counterpart in not increasing the length of a
formal object, it does need some "add-on" linking mechanism to make the
(unselective) binding relation work. The analysis represented by (66a) thus
appears to follow from a richer theory which requires unselective binding in
addition to LF wh-movement:

(66) a. Ix"A lx' ...


wh...l] -*[x"Op[o] [x.... wh...J
-I[x.Opi(l] [x... wh(i) ... ]B

wh...]-- [x" whi [x' ...


b. [x" A [x' ... ti...]

Conseqently, (66a,b) may not be subject to comparison on the ground of


Economy, even if some languages do prefer (66a) to (66b).
This observation has inspired us to look further into various types of
unselective binding construal on different levels in different languages. The
result can be visualized in the following diagram (word orders irrelevant):

(67) CP/IP

Vx /doux ..... PP/DP


(Chinese)
kax/mox ..... Xo
(Japanese)
somex /everx wh(x)
(English)

48
Our observation, though by no means infallible, seems to lead to an interesting
conclusion: Unselective binding is an augmentation of lexical binding (or, in a
sense, an equal of pronominal binding). In other words, the relation between
shei 'who' and dou 'all' in Chinese, as well as that between dare 'who' and -mo
'all' in Japanese, is no different from the relation between who and -ever in
English. Under this view, the distinction between external and internal binding
in Nishigauchi's (1990) sense is only an illusion, created by morphological
makeups of individual languages and imposed by principles associated with
particular configurations (e.g., the revised PCOB; cf. section 2.3).
We thus expect that interrogative construals also display the same trait
across languages. First let's consider Japanese wh-questions. Although it is still
under debate whether Japanese displays genuine wh-island effects (cf. Lasnik
& Saito 1984,1992 vs. Nishigauchi 1986,1990, Pesetsky 1987, and Watanabe
1991), it is generally agreed that Japanese lacks Complex NP (i.e., strong
island) effects. Let's assume that there does exist such an asymmetry between
Complex NPs and wh-islands in Japanese, and see what our theory may say
about it. The solution turns out to be quite straightfoward, as illustrated below
(data from Watanabe 1991, categorial labels attached to the right in Japanese
for ease of exposition);

49
(68) a.??[[[John-wa [[Mary-ga [[nanii NPI ti DP]-o kattalp] ka-dooka cP]
John-Top Mary-Nom what -Acc bought whether
Tom-ni tazuneta ip] no c] OpiOp] CP] ?
Tom-Dat asked Q
What is the thing x such that John asked Tom whether Mary bought x?
b.??[cp Whati [a did [ip John ask Tom [cP whether [IP Mary bought ti ]]]]] ?

Following Watanabe (1991), we may assume that a Q-operator originates from


the Spec of DP in Japanese, as in (68a). Consequently, it patterns with English
wh's (i.e., full wh-phrases containing a Q-operator, cf. section 2.1) in displaying
wh-island effects, as in (68b). On the other hand, since the Q-operator is already
in the Spec of DP, any wh-in-situ within its c-command domain (and hence
within a Complex NP) can be licensed by unselective binding, as illustrated
below:

(69) [[[John-wa [[[[tk dare(x) -o aisiteiru p]J Opk CP] onnak NP] t1 DP]-o
John-Top who-Acc loves woman -Acc
nagutta Ip] no c] Opx [a] c] ?
hit Q
Who is the person x such that John hit the woman who loves x?

This is exactly what we would expect in view of the lack of locality effects in mo-
quantification, as exemplified below ((48b) repeated here):

50
(70) [[[dare(x) -ga tk kai-ta IP] Opk CP] tegamik Dp]-ni-mox pp] onazi
who-Nom wrote letter -in-all same
kota-ga kai-te-at-ta.
thing-Nom written-was
For all x,y, x a person, y a letter x wrote, the same thing was written in y.

This analysis of the lack of strong island effects in Japanese carries over to
Chinese directly, except that the position of Chinese binders is probably much
higher than PPs or DPs, as mentioned above. As evidenced by the parallel
between the long-distance interrogative construal in (71) and dou-quantification
in (72), this class of A'-dependency does observe Subjacency or the ECP:

(71) [cP Opx j[o [IP [DP [CP Akiu de piping [pp dui shei(x) ]zaocheng]
Akiu PNM criticism about who cause
de shanghai] zui da]]?
PNM damage most great
Who is the person x such that [the damage [which Akiu's criticism
caused x]] is greatest?
(72) [DP [CP Akiu de piping [Fp,dui shei(x) ] zaocheng] de shanghai]
Akiu PNM criticism about who cause PNM damage
doux henda.
all great
For every x, x a person, [the damage [which Akiu's criticism caused x]]
is great.

The difference between Chinese and Japanese therefore lies in the fact that the
former clearly lacks wh-islands effects, as evidenced by the following classic
example from Huang (1982):

51
(73) ni xiang-zhidao [shei mai-le shenme] (ne/ma)?
you want-know who buy-Prf what Qwh/Qyes/no
a. Who is the person x such that you wonder what x bought? (Qh)
b. What is the thing x such that you wonder who bought x? (QwI)
c. Do you wonder who bought what? (Qyesino)

For skeptics who question the wide scope question construals of (73), we may
further consider the following examples. The idea is to single out the wide
scope readings by invoking the incompatibility between wh-questions and
yes/no-questions:

(74) ni xiang-zhidao [shei lai-bu-lai] (ne)?


you want-know who come-not-come Qwh
a. Who is the person x such that you wonder whether x will come?
b. #Do you wonder who will come?

(75) ni xiang-zhidao [Akiu mai-bu-mai shenme] (ne)?


you want-know Akiu buy-not-buy what QMh
a. What is the thing x such that you wonder whether Akiu will buy x ?
b. #Do you wonder what Akiu will buy?

The judgement is clear-cut. Native speakers who usually have trouble in


processing (73) pick up the wide scope readings without much difficulty.
The distribution of locality effects in the three types of language may thus
be summarized as follows:

52
(76) English Japanese Chinese
Wh-island effects yes yes no
Complex NP effects yes no no

This is exactly what we would expect from the distinct positions of Q-operators
in these languages, as illustrated below (word orders irrelevant):

(77) a. Chinese-type: [cp Opx [a] [IP ... wh(x) ... I]

b. Japanese-type: [cP Opx [0] [IP ... [pp/op9 tx [ ... wh(x)...] ... ]

c. English-type: [cP [PP/DP wh(x)-OPx [o01]k][IjP ... tk... ]]

Since Chinese Q-operators are inserted in the CP Spec, no movement is


involved. In contrast, since Japanese Q-operators are inserted in the DP Spec,
the "half-way" movement to the CP Spec evades Complex NP effects, but still
respects the wh-island constraint, or whatever principle it might be reduced to.
As for English, since the whole wh-phrase must move to fulfill feature-checking
on the CP Spec, both Complex NP and wh-island contraints are to be observed.
If our analysis proves to be on the right track, then the linking mechanism
required in (66a) is only an ,Annotation of the general binding relationship
behind any operator-variable pair, definable by the notion of c-command. Since
the relationship can be realized in such a minimal way, it manifests itself
maximally across languages. Along this line, the basic intuition behind the LCH
seems to hold, and the intrinsic priority of (66a) over (66b) can still be made to
follow from a broader notion of Economy.

53
3. Nouns vs. Adverbs

As a reflection, we have been pursuing an ideal design at the cost of a


popular assumption; that is, all wh-phrases are created equal. As a matter of
fact, it is crucial for our purpose here that wh-phrases vary in their internal
structures, not only across languages, but also across categories. Although it is
not clear so far that this is indeed the case, 14 our analysis is essentially in line
with the spirit of the minimalist approach; namely, languages differ only in the
lexicon and PF.
In the same vein, our next proposal is to follow Tsai (1992c,1993) in
claiming that the argument-adjunct asymmetry in question is essentially a noun-
adverb asymmetry under the assumption that only nouns may introduce pure
(i.e., [-pronominal]) variables in situ. 15 This move captures the essence of
Higginbotham's (1983,1985) proposal that N is generated with an index-
argument, which must be "discharged" in terms of binding from a determiner.
For instance, a definite DP such as the donkey is analyzed as an operator-

14 See Cheng (1991) for a cross-lingiistic survey of the architectures of wh-


phrases and wh-particles (i.e., Co[+wh]), which seems to be consistent with our
conjecture. Particularly of our interest here is the descriptive generalization that
if a language has a wh-particle, the language always uses it (cf. Cheng
1991:28). We will return to this issue later.
15 The possibility of distinguishing nominal wh-porases from non-nominal
ones in regard to their extraction behavior has actually been entertained by
Huang (1982) along with the argument-adjunct distinction. He argues that
where and when are actually nouns (vs. adverbs such as how and why), and
hence their patterning with arguments in regard to LF locality effects. Based on
Huang's initiative, Tsai (1991) further explores the distinction in terms of the
generalized binding theory (Aoun 1985, 1986, WAHL 1987), though still within
the tradition of "all-out" movement analyses.
54
variable pair based on its internal structure (78):

(78) a. DP

Det NP
I I
the N
I
donkey (i)
b. thex (donkey (x))

(78a) can then be mapped straightforwardly into the usual logical


representation for the DP, i.e., (78b). Reinhart (1992, 1993) extends this plot
further to which-NPs, with a view to deriving the well-known asymmetry
between them and wh-pronominals in regard to superiority and/or crossing
effects (cf. Pesetsky 1987):

(79) a. DP

Det NP
I I
which N
I
donkey (i)
b. f { xldonkey (x)}

Here which is treated as a weak determiner, which by definition is defective in


serving as an operator. As an alternative, Reinhart suggests that together with a
set introduced as the translation of N, it forms a (choice-) function variable, as in
(79b). As a result, which-NP may be interpreted in situ given Baker's (1970) Q-
morpheme hypothesis (see also Katz & Postal 1964, Pesetsky 1987). In

55
contrast, who and what are categorized as determiners, which project directly to
DP:

(80) DP

Det
I
who/what

The difference is that although who and what may in theory be translated as
functions, there is no N-set in (80) for them to apply to. Consequently, the only
way to realize their quantificational force is to undergo wh-movement. In the
light of the structural distinction between wh-pronominals and which-NPs,
Reinhart (1992) proposes that which man in (82b) does not undergo LF
movement, since it introduces an N-set, and is therefore eligible for unselective
binding from the matrix Q-morpheme:

(81) a. Whoi did you persuade ti to read what?


b.??Whatj did you persuade whom to read tj?

(82) a. [comp Q<i,j> which mani ] did you persuade tj to read which book(j) ?
b. [comp Q<i,jp which bnokj ] did you persuade which man(i) to read tj ?

In contrast, its pronominal counterpart in (i.e, whom) has to move in LF, and
hence the crossing effects displayed by (81b). Her treatment thus captures

Baker-Pesetsky's insight without resorting to the notion of D(iscourse)-linking.


On the other hand, we are not to dismiss the intuition that a which-NP
tends to be D-linked (cf. Pesetsky 1987). From our point of view. the D-linking

56
effect is essentially a result rather than the cause of the asymmetry between
which-NPs and wh-pronominals. Namely, when which does function as a strong
determiner (i.e., undergoing wh-movement), the head noun of (79a) readily
provides a restrictive clause, inducing the presuppositional (D-linked) reading
(also cf. Heim (class lecture), Kroch 1989, Diesing 1991). Under this view, what
is responsible for the D-linkedness of (82a,b) is the fronted which-phrase rather
than the in-situ one. As a result, we not only come very close to Cinque's (1989)
original conception of referentiality and its connection with nominality, but also
capture the crobs-linguistic generalization that which-questions typically come
with a pre-established set of answers.
One draw back here is that neither the structural solution nor the D-linking
account provides a satisfactory explanation of the long-distance construal in
multiple wh-questions. First consider the following contrast:

(83) a. Who remembers [where we bought what]?


b.??Whati does John remember [where we bought ti ]?

It has been widely noted that the wide-scope reading of what in (83a) is
licensed by the presence of another wh-phrase in the main clause. If what has
to undergo LF movement, as it is a wh-pronominal and/or a non-D-linked wh-
phrase, we would expect the same locality effect displayed by (83b). This
prediction, as evidenced by (83a), is not borne out. Consequently, we are led
back to exactly where we began, i.e., the postulation that Subjacency does not
hold in LF.
A possible way out is to follow Reinhart's (1993) refinement, replacing (80)

57
with a structure parallel to (79a), i.e., [DP who/what [N' e(i) ]]. As a matter of fact,

this analysis of wh-pronominals is virtually equivalent to those represented by


(17) and (19), except, that it is conceived in terms of phrasal syntax. She
proposes to derive the crossing effect of (81b) under Economy considerations:
Since it is always the c-commanding wh-phrase which crosses fewer nodes, the
derivation of (81a) is less costly than that of (81b), although it still remains to be
seen why (82a) does not block (82b) under the same consideration.
We can then account for the lack of locality effects in (83a) by recasting the
Baker-Pesetsky's analysis in terms of the IP-CP system, as illustrated by the
following LF representation:

(84) [cP whoi [c Q<i,k> [IP t- remembers [cP wherej [a Q<O> [IP we bought
what(k) tj ]]fl]]?

Under this analysis, the relevant unselective binder in (84) is the matrix Q-
Comp, which remains inactive until being "turned on" by Spec-head agreement
with the moved-in wh-phrase (that is, by morphological checking in Chomsky's
(1992) sense). Since the wide-scope construal of what in (84) is achieved by
unselective binding instead of Chain formation, no Subjacency and/or
relativized minimality violation is expected. This move, in a sense, also provides
a more explicit mechanism for the Scope Absorption analysis (Chomsky 1986b,
1992).16

16 The notion of "absorption" is first defined by Higginbotham & May (1981) to


explain the "paired" reading of multiple wh-questions. The intuitive idea is that
(assuming the S-S' system,) a number of wh-phrases can oe packed together
into an n-ary operator at a Comp site, as illustrated below:
58
Along the line sketched above, we then expect Chinese wh-phrases to
pattern with which-NPs (andwh-pronominals in the new light) in terms of their
status as (function) variables. This possibility has already been explored by
Cheng (1991) and Li (1992) with fruitful results: Chinese wt-phrases appear to
lack their own quantificational properties, and behave in line with polarity items
in Klima's (1964) sense. (See also Huang (1982:241-253) for original
discussions of indefinte wh-construals under negation, A-not-A questions,
conditionals, and dou-quantification).
Our prediction, however, is not entirely borne out. As we have seen in (4),
wh-adjuncts such as weishenme 'why' apparently do not fit into the picture.
They display both strong and weak island effects when wide-scope question
formation is involved, and in general resist indefinite construals. For example,
while shenme 'what' can be embedded within conditionals and read as
'something', as in (85), the same construal is impossible for weishenme, as
evidence by (86):

(85) ruguo Akiu mai-le shenme, ta yiding hui lai gaosu wo.
if Akiu buy-Prf what he surely will come tell me
If Akiu bought something, he surely will come to tell me.

[Comp Whl,Wh2,..., Whn ][s...1]]--


is'
(i) , n}] [s . ]]
* s'[Com Wh{1,2,....
Although it has been suggested by Chomsky (1992) that this analysis can be
adopted in a rather loose form to account for (83a), in that what is absorbed (or
"attracted" in Chomsky's (1986b:53) terms) by another wh-phrase in a scope
position, it is not entirely clear how this generalization can be properly
formulated, and whether there is a genuine connection between the two
operations. For instance, the absorption rule (i) may be independently needed
in multiple wh-fronting languages (cf. Rudin 1988), while "scope absorption"
from an abstract Q-operator (or a lexical Q-marker, if any) is a standard practice
in Chinese-type languages.
59
(86) * ruguo Akiu weishenme buneng jiao zuoye,
if Akiu why
tayiding
hui
cannot hand-in homework he surely will
lai gaosu wo.
come tell me
If for some reason Akiu cannot hand in homework, he surely will come
tell me.

Nevertheless, the asymmetry is not so surprising if again we put English


wh-words under the microscope: While we can easily pick out some-what, what-
ever, or even what-so-ever in the dictionary, (non-)words like *some-why and
*why-ever are never to be found. 17 This in turn suggests that why, unlike who
and what, is not subject to binding construals, as illustrated below:

(87) a. * Advo b. * Advo

some-x Advo AdvO -everx

wh- y(x) wh- y(x)

Nor does Japanese allow universal quantification over naze 'why': *naze-mo,
which would mean 'for any/every reason' if well-formed, is not a possible
combination.18

17 Here we systematically leave out compounds such as how-ever and some-


how, which do not fall into the same category as why. We will defer the relevant
discussion to chapter two, where the peculiar properties of how and its Chinese
counterpart zenmeyang will be re-examined.
18 For some reason, Japanese does allow naze-ka 'for some reason'. We will
take it to be an isolated case for the time being. On the other hand, however and
somehow do have counterparts in Japanese, i.e., doo-mo 'by any/every means'
and doo-ka 'by some means'. See chapter 2 for further discussion.
60
All in all, we find that it is quite safe to assume that adverbs, as intrinsic
operators, do not enter into unselective binding as variables. Rather, being

denied access to binary substitution, they appeal to Move-a to avoid vacuous

quantification. The traces which they leave behind then count as variables for
binding purpose. Given Huang's (1982) LF movement analysis, we thus have a
principled account of why (4) displays ECP/Subjacency effects, and why LF
representations like (6) are impossible.
The same analysis applies to the contrast between (85) and (86). As
observed by Cheng & Huang (1993), the existential construal of the wh-in-situ

in (85) results from existential closure (3-closure) triggered by ruguo 'if', given

that Chinese wh-phrases count as polarity items (cf. Cheng 1991, Li 1992), as
illustrated by (88):19

(88) [cp ruguo 3x[IP Akiu [vP mai-le shenme(x)]]], ....


if Akiu buy-Prf what

Here a technical problem has to be solved before we proceed. For typical

19 Note that multiple wh's-in-situ are possible in conditionals, as exemplified


by (ia):
(i) a. ruguo shei gei-le shei shenme, ni jiu lai gaosu wo.
if who give-Prf who what you then come tell me
Ifsomeone gives someone something, then you come to tell me.
b. [cP ruguo 3x,y,z [Ipshei(x) gei-le shei(y) shenme(z) ]]],....
if who give-Prf who what
There are altogether three wh-phrases in the conditional clause, and all of them
are construed as indefinites. It is therefore natural to assume that they are
bound by an unselective binder (i.e., 3-closure on the IP node), as illustrated by
(ib).
61
indefinites like yi-ge ren 'a person', 3-closure appears to stick to VP (cf. Diesing

1992) or syntactic predicates (cf. chapter 3), as evidenced by the obligatory


presence of you 'have' in (89), which contributes existential force to the subject
indefinite:

(89) *(you) yi-ge ren (3)[vP lai-le].


have one-CL person come-Prf
One man came.

This holds for Chinese wh's-in-situ when the trigger in question occurs between
IP and VP, as in the case of negation and probability operators:

(90) Akiu bu 3x [vP yao shenme(x)].


Akiu not want what
a. Akiu does not want anything.
b. What doesn't Akiu want?

(91) Akiu dagai/keneng 3x [VP yao shenme(x)].


Akiu probably/possibly want what
a. Akiu probably/possibly wants something.
b. What does Akiu probably/possibly want?

As shown above, the object wh's-in-situ can be construed as either indefinite or


interrogative (see chapter 3 for detailed discussion). On the other hand, these
triggers do not license subject wh's-in-situ in the same configurations, as
evidenced by the lack of indefinite readings in the following examples:

62
(92) shei bu yao kafei?
who not want coffee
a. #Someone/Anyone does not want coffee.
b. Who does not want coffee?

(93) shei dagailkeneng yao kafei?


who probably/possibly want coffee
a. #Someone probably/possibly wants coffee.
b. Who probably/possibly wants coffee?

This indicates that, as far as polarity items are concerned, the scope of 3-

closure is determined by the structural position of its trigger. Our observation is


further supported by the fact that a subject wh-in-situ does get licensed within
ruguo-conditionals, incontrast to its counterparts in (92) and (93):

(94) ruguo 3,x [p shei(x) mai-le chezi], Akiu yiding hui lai gaosu wo.
if who buy-Prf car Akiu surely will come tell me
If someone bought a car, Akiu surely will come to tell me.

Consequently, a line has to be drawn between Chinese indefinites and wh's-in-

situ with respect to the domain of 3-closure.

With this knowledge in mind, we may account for the deviance of (86) in a
straightforward manner. On the one hand, weishenme 'why' cannot be targeted

by unselective binding from 3-closure on the IP node, as illustrated below:

(95) * [cP ruguo 3x [Ip Akiu weishenme(x) bu-neng [vP jiao zuoye]]], ....
if Akiu why can-not hand-in homework

63
On the other hand, since there is no intermediate scope position for wh-phrases
in (95), weishenme have to move all the way to the matrix CP Spec. As a result,
(86) is ruled out by Subjacency, the ECP, and possibly Relativized Minimality.

64
4. Strong vs. Weak Operator Features

An issue which we have not touched upon so far concerns the status of
Chinese Comp in regard to feature checking; namely, whether its operator
feature is strong or not. We did not concern ourselves with this aspect mainly
because, given that the design (8a) is realized in an IP-CP magnitude, the Q-
operator in question should satisfy the checking requirement vacuously.
Nonetheless, since we have introduced the noun-adverb distinction in terms of
their ability to enter into binding construals, it becomes necessary to spell out
the inevitable: Operator features are weak in Chinese, and the procrastination
principle applies accordingly to block overt wh-extraction. In other words,
Chomsky's (1992) generalization that operator features are universally strong is
probably too strong for our purposes here.
For one thing, there does not seem to be an a priori reason why languages
should agree on the status of operator features, while they diverge with respect
to N-features and V-features, as generally assumed to be the case with English
and French head movement (cf. Chomsky 1991). The issue, as it turns out, is
really an empirical one. So let's start with the null hypothesis that languages
vary in regard to the strong/weak status of their operator features, and see how
far it can go.
First note that if operator features are strong across languages, the claim
that Subjacency holds only in overt Sy. mxserves as an adequate
still
descriptive generalization. Our task, therefore, is to find out if there exists a
language where wh-phrases are in general allowed to stay in situ, but

65
nonetheless display Subjacency effects. Hindi wh-questions seem to be a
promising candidate, as evidenced by (96a,b) (All Hindi data below from
Mahajan 1990):

(96) a.* raam-ne socaa ki [DPyah baat [cp ki mohan-ne kis-ko maaraa] galat hE
Ram-erg thought this fact that Mohan-erg who hit wrong is
Who is the person x such that Ram thought that [the fact [that Mohan
hit x]] is wrong?
b.* raam-ne [DPUS aadmii-ko [cp jis-ko ravii-ne kyaa ciiz dii 1]] baazaar
Ram-erg that man who Ravi-erg what thing gave market
jaate dekhaa
going saw
What is the thing x such that Ram saw [the man [who Ravi gave x ]]
going to the market?

As Mahajan points out, the deviance in question poses non-trivial problems for
S-structure Subjacency. He then proposes that Hindi wh-phrases do not
undergo LF wh-movement; rather, they are quantifier-raised (QRed) to adjoin to
the immediately dominating IPs (see also Kim 1991). By restating the Wh-
Criterion in terms of government (originally defined in terms of dominance in
May (1985)), (96a,b) is then ruled out by selectional restrictions on the matrix
[+wh] Comp, in that the government relation between CO[,Hw and the wh-phrase

in question is blocked by the DP and CF node, both barriers in Chomsky's


(1986b) sense (see also Nishigauchi (1986) for a similar treatment in terms of
wh-feature percolation). Consequently, there is no need to reject S-structure
Subjacency under Mahajan's analysis, since no movement is involved in

66
assigning scope to the adjoined wh-phrase.
This approach, though plausible in its own right, is incompatible with the
checking mechanism developed by Chomsky (1992), where selectional
-estrictions are satisfied in a Spec-head configuration. Nor does it fit into Rizzi's
(1992) formulation of Wh-Criterion (cf. (10)) for exactly the same reason. A
simple way out, as we might expect from the discussion at the beginning of this
section, is to follow Tsai (1992b) in claiming that operator features are weak in
Hindi. Consequently, overt wh-movement is blocked by the principle of

procrastination. When Move-a does apply in LF, it induces a strong Subjacency

violation since two barriers (i.e., DP and CP) are crossed. We then have a
compatible account within the minimalist framework.
Another related fact comes from CED effects associated with extraposed
complements: As Mahajan (1990) observes, a finite clausal complement always
extraposes in Hind (and hence is located to the right of the main verb), t:nd an
extraposed clause does not allow an in-situ wh-phrase, as shown by (97a). To
make the question licit, the wh-phrase must undergo overt fronting, as shown by
(97b). In contrast, when a clausal complement does not extrapose (i.e.,
remaining to the left of the main verb), as is the case of infinitivals like (98),
wh's-in-situ are allowed:

67
(97) a. * raam-ne ti socaa [cP ki mohan-ne kis-ko dekhaa]i
Ram-erg thought that Mohan-erg who saw
Who did Ram think Mohan saw?
b. kis-koj raam-ne t socaa [CP ki mohan-ne tj dekhaa]i
who Ram-erg thought that Mohan-erg saw
Who did Ram think Mohan saw?
(98) raam-ne [PRO kis-ko dekhnaa] caahaa
Ram-erg who to see want
Who did Ram want to see?

First consider (97b). Since the extraposed clause is an ungoverned domain, it is


unlikely for the embedded subject kis-ko 'who' to move after extraposition, in
violation of the CED (Huang 1982) or Subjacency (Chomsky 1986b). Therefore,
wh-fronting must have preceded extraposition in (97b).20 In case wh-fronting
does not apply at all, as in (97a), the wide-scope construal is blocked. This
deviarnce is totally expected, because, under our approach, Subjacency works
in exactly the same way in LF as it does in overt Syntax. The delayed LF
movement in (97a) is thus ruled out along with that in (96a) (also cf. Srivastav
1991). On the other hand, since the infinitive complement is still L-marked by
the verb caahaa 'want' in (98) (recall that extraposition of infinitivals is not

20 Mahajan (1993) provides further evidence to his conclusion here. He


points out that it is possible to replace the CP trace in (97b) with an expletive
yah 'it', as exemplified by (i), in which case no extraposition is involved:
(i) ???kis-koj raam-ne yahi socaa [cpki mohan-ne tj dekhaa]i
who Ram.erg IT thought that Mohan-erg saw
Who did Ram think Mhohan saw?
Consequently, there is no way for kis-ko 'who' to escape from the
CED/Subjacency interms of overt fronting, and hence the deviance of (i).
68
obligatory), LF movement does not induce any CED effect. The procrastination
analysis thus makes the right prediction again.
The story, however, does not quite end here. A notable problem with
Watanabe's (1991) analysis with respect to Hindi wh-auestions concerns cases
like (97a). Here we expect that the wh-operator generated in the DP Spec of
kis-ko moves in overt Syntax, and thereby escapes the CED effect in exactly the
same Tashion as the full wh-phrase does in (97b).21 Nevertheless, as Masa
Koizumi (p.c.) points out, this will not be a necessary conclusion if we assume
that (pure) wh-operators are not subject to scrambling, which is a common
practice for tall wh-phrases in both Hindi and Japanese. Under this view, the
wh-operator has to move all the way to the local CP Spec before extraposition
applies, to avoid a CED violation. When extraposition does apply, presumably
adjoining the wh-ph.ase to the right "roof" of IP, strict cyclicity is violated, since
the IP-adjunction in question does not extend its target, i.e., the matrix CP (cf.
Chomsky 1992), as illustrated below:

(99) [cP Opi ([][IP raam-ne [cP ki mohan-ne [DP ti [NP kis-ko]j dekhaa] socaa]]
Ram-erg that Mohan-erg who saw thought

-X- [cP Opi [l [IP[IP raam-ne tk socaa] [cP ki mohan-ne [DP ti [NP kis-ko]]

dekhaa]k ]]

(97a) is thus ruled out correctly. On the other hand, since the whole wh-DP in
(97b) has been scrambled out of the finite complement before extraposition, the

21 The same problem is independently noted by Mahajan (1993), who rejects


the pure wh-operator analysis from quite a different angle.
69
so-called "invisible" wh-movement (i.e., movement of the pure wh-operator) may
be postponed until extraposition applies, and thereby observe strict cyclicity.
The derivation is given as follows:

(100) [CP [IP [DP Op[o] [NP kis-ko]]i raam-ne [cP ki mohan-ne tj dekhaa] socaa]]
who Ram-erg that Mohan-erg saw thought
-[CP [IP [IP [DP Op[Q] [NP kis-ko]]i raam-ne tk socaa] [cP ki mohan-ne ti
dekhaa]k ]]
-- [cP Opji [l [IP [IP [DP tj [NP kis-ko]]i raam-ne tk socaa] [cP ki mohan-ne ti
dekhaa]k ]]

As noted by Koizumi, there is still a technical problem to tackle in this solution:


The wh-DP, when scrambled, becomes an ungoverned domain, and
supposedly constitutes an island for the ensuring invisible wh-movement
according to the CED. The same observation applies to Japanese scrambling
under strict cyclicity. Namely, scrambling cannot occur after invisible wh-
movement. As a result, scrambling has to be undone in L.F (cf. Saito 1989). This
move, however, renders invisible wh-movement an LF operation, and results in
a dilemma. We will leave the issue open here.
Now let's take a step back from the technical details, and look at the big
picture. First, as confirmed by Mahajan (p.c.), Hindi does not have the parallel
morphology which Japanese employs in licensing indefinite wh's (cf. '(uroda
1965, Nishigauchi 1986, 1990, Watanabe 1991, among others), which is one of
the major motivations for Watanabe's proposal. The null hypo hesis, therefore,
is that Hindi wh-phrases move as a whole due to their morphological makeups.
The burden of proof is thus shifted to the pure wh-operator hypothesis: Unless

70
there exists evidence in Hindi showing that the "invisible" movement analysis
explains something which the "covert" movement analysis cannot, our position
is defenldable.
Second, Hindi and Japanese also differ with regard to strong island
effects. Namely, Japanese allows long-distance construals of wh-arguments
embedded in complex NPs. This point can be made clear by comparing
(101a,b) with (96a,b) respectively (Data from Lasnik & Saito 1992):

(101) a. kimi-wa [DP [CP Taroo-ga nani-o te-ni ireta] koto]-o sonnani
you-top Taroo-nom what-acc obtained fact-acc so mu;.i
okotteru no?
angry Q
What is the thing x such that you are so angry about [the fact [that
Taro obtained x ]]?

b. kimi-wa [DP [CP ei nani-o katta] hitoi ]-o sagasite iru no?
you-top what-acc bought person-acclooking-for Q
What is the thing x suc t.hat you are looking for [the person [who
bought x]]?

Watanabe (1991) observes that the lack of locality effects in (101a,b) can be
explained if we assume that there is a wh-operator located in the DP Spec of
the entire complex NP, which serves as a binder of the wh-in-situ down below
(i.e., nani 'what'). Consequently, if the strict cyclicity account goes through (that
is, if wh-operators indeed cannot scramble), it still remains to be seen why
Japanese allows insertion of a wh-operator in the topmost Spec of a complex
NP, while Hindi does not,

71
Finally, there is a potential paradox between the strict cyclicity account and
the extension of the pure wh-operator analysis to kyaa-questions. If we are to
identify kyaa as an overt wh-operator in Hindi, then the analysis presented by
(98) cannot be carried over to examples like (102), where kyaa appears to
undergo successive cyclic movement (i.e., leaving overt copies in intermediate
adjunction sites) before extraposition, and escape from the CED effect:

(102) raam-ne kyaa socaa [cP ki ravil-ne kyaa kahaa [cP ki


Ram-erg KYAA thought that Ravil-erg KYAA said that
kCn sa aadmii aayaa thaa]].
which man come
Which is the man x such that Ram think [that Ravi say [that x came]]?

It is also instructive to note that the extension cannot hold if Srivastav


Dayal (1993) is correct about her indirect dependency analysis, where kyaa is
related to kOn sa aadmii 'which man' through semantic composition of two local
wh-dependencies. 22 Under this approach, kyaa is treated syntactically as an
expletive in an argument position, linked to an "extraposed" indirect question,
and semantically as a scope marker, quantifying over propositions (i.e.,
possible answers). More importantly for us, kyaa undergoes LF movement to
the matrix CP Spec to fulfill feature checking. Her view thus neatly dovetails our
analysis of Hindi question formation.

22 See also Mahajan (1993) for arguments against generalizing the pure wh-
operator analysis to kyaa-questions.
72
5. An Overview

So far we have sketched three basic proposals: First, binary substitution


has intrinsic priority over singulary substitution. Second, only nominals, but not
adverbs, are subject to binding. Third, languages vary in regard to the
strong/weak status of their operator features. Now we can see how these three
could combine to derive the linguistic variations at issue, not only across
languages but also across categories. Putting the last thing first, we may group
Hindi and Chinese together in terms of the status of their operator features:
Since operator features are weak in these two languages, wh-movement
procrastinates until LF. In contrast, their English and Japanese counterparts are
strong (though the observation is still subject to debate on the part of
Japanese). And henre overt movement of wh-phrases in English and that of
empty wh-operators in Japanese, as illustrated below:

(103) English Japanese Chinese Hindi


- - - - - - - - - - - - - - - - - - - - - - - - -II--- - -I

Operator features strong (?) strong weak weak


LF wh-movement no (?) no yes yes

On the other hand, as we have demonstrated throughout section 1,


Chinese and Japanese should be grouped together with respect to their
relatively high maneuverability of (unselective) binding. Namely, both Japanese
and Chinese allow long-distance indefinite construals of wh's, as exemplified in
(de)mo-conditionals and dou-conditionals respectively (cf. section 2.4), and
only Chinese allows bare conditionals (cf. section 2.3). Ve may thus employ the

73
following criterion to capture the above intuition:

(104) The Watchtower Criterion:


a. The higher a binder is, the farther the binding will be.
b. The farther a binding is, the higher the binder should be.

(104) therefore provides a rationale for the descriptive generalization which we


have been depicting in the last tree sections, as illustrated in the following table:

(105) English Japanese Chinese Hindi

Non-quantificational wh's no yes yes no


(De)mo/dou-conditior.als (?) no yes yes (?) no
Bare conditionals no no yes no
Binding construals of lexical phrasal sentential lexical
operator-variable pairs

It is instructive to note that (105) only provides a general outline of patterning


and contrasting, and there do exist some overlaps among these languages. For
intance, English no matter-constructions bear partial resemblance to (de)mo-
and dou-quantification, in that they only license E-type pronoun construals in
consquent clauses. The difference is that (de)mo and dou may take scope over
either DPs or CPs (hence qantifying over either individuals or propositions),
while no matter only takes scope over CPs, quantifying over propositions or
situations. We will return later to elaborate on the syntactic aspects of this
distinction.
The interaction between the above two sets of factors thus produces the
now familiar pattern of interrogative construals in regard to locality effects:

74
(106) Single Wh-questions English Japanese Chinese Hindi
Wh-island effects yes (?)yes no yes
Complex NP effects yes no no yes

Since English wh-phrases are neither subject to procrastination nor subject to


indefinite construals above the XO level, overt wh-movement applies and locality
conditions are observed strictly. Hindi follows similar patterns except that wh-
movement applies in LF due to the weak status of its operator features. Like
Hindi, Chinese wh-movement applies only in LF. But unlike Hindi, it applies
only on the part of wh-adverbs, which cannot introduce variables in-situ and
therefore are not subject to binding. Japanese, on the other hand, displays
mixed behavior, because of the structural position of its Q-operator (i.e., the
Spec of DP/PP), which allows the evasion of Comlex NP islands, but not wh-
islands.
Furthermore, the Lexical Courtesy Hypothesis (2) also sheds new light on
the way we look at cross-linguistic variation. That is, languages may appear to
specialize in some specific type of operation (hence the traditional distinction
between syntactic and abstract wh-movement languages). Nonetheless, they all
share the basic design as to how an operator-variable pair should be
constructed, and only diverge because of the tension between Syntax and
Morphology, the logical necessity imposed by particular configurations, and the
idiosyncracies resulted from historical changes and regional influences.

75
CHAPTER TWO

SOME ASYMMETRIES BETWEEN CHAIN FORMATION

AND UNSELECTIVE BINDING

1. Long-Distance Construals ofAmount Wh's

So far we have demonstrated that there is a principled way to determine


where and when unselective binding applies with respect to a fairly restricted
set of factors, resulting in a noun-adiverb asymmetry.What we haven't shown is
whether this move is to characterize a distinct type of A'-dependency, or just to
provide an alternative taxonomy of wh-construals. To begin with, Chain
formation can be further divided into two categories, i.e., successive cyclic
movement and long movement (Cinque 1989,1990, Rizzi 1990), with the latter
being understood as an instance of long-distance binding (or individual
variable binding in the sense of Heim (1987) and Frampton (1990)).1 The
hallmark of long movement construals is that they display Subjacency/CED
effects, but do not observe the (conjunctive) ECP and Relativized Minimality.
This point is illustrated by the following contrast:

(1) a. ?[How many fish]i do you wonder whether John weighed t i ?


b. *[How many pounds]i do you wonder whether John weighed t i ?

1 See also Cresti (1994) for a more elaborated view along the line of individual-
variable accounts.
76
(2) a. [How many fish]i do you think John weighed t i ?
b. [How many pounds]i do you think John weighed t i ?

As Cinque (1989) points out, although both how many fish in (la) and how

many pounds in (1b) are 0-marked by the verb weigh, only the former, but not

the latter, escapes from the ECP effect. This point can be made clear by
comparing the wh-island constructions (la,b) with the bridge-verb constructions
(2a,b) respectively.
The reason, as provided by Cinque, is that the amount wh-phrase how
many pounds is not referential, and therefore not eligible for binding construals
(see also Koopman & Sportiche 1988, Rizzi 1990). As a result, it must undergo
successive cyclic movement, and accordingly observe both the ECP and
Subjacency. !n contrast, only referential expressions such as how many fish
may undergo long wh-movement, by definition immune to the antecedent-
government requirement.
On the other hand, our characterization of unselective binding dictates an
absolute absence of Subjacency and ECP effects, as long as wh phrases
involved introduce variables in situ. This prediction is indeed borne out by the
following wide scope construal of Chinese amount wh-nominals:

(3) ni xiang-zhidao [shei zhong duoshaoli bang] ne?


you want-know who weigh how-many pound w
a. Who is the person x such that you wonder how many pounds x weighs?
b. What is the number/amount x such that you wonder who weighs x
pounds?

77
As evidenced by (3b), duoshao bang and ji bang pattern with argument NPs in
not displaying wh-island effects in wide-scope question formation (cf. Huang
1982). It is also instructive to note that zhong 'weigh' is construed only as stative
(vs. agentive) in Chinese, as evidenced by (4a):

(4) a. *Akiu zhong liang-tiao yu.


Akiu weigh two-CL fish
Akiu weighs two fishes.
b. Akiu zhong *(Iiang-bai bang).
Akiu weigh two-hundred pound
Akiu weighs two hundred pounds.

(4a) further indicates that the amount NP in question is lexically selected. As


pointed out by Howard Lasnik, the same pattern applies to English multiple wh-
questions:

(5) a. *How many pounds does [who remember whether John weighed t ]?
b. Who remembers [whether John weighed how many pounds]?

Since there is no referential DP (or DP ranging over individuals) such as how


many fish involved in (3b) and (5b), it is clear that this lack of wh-island effects
in LF calls for independent treatment other than the long movement analysis.
One might suspect that the bond between zhong and duoshaolJi bang in
(3) is much stronger than that between their English counterparts in (lb): It is not

impossible that zhong both lexically selects and 0-marks duoshao/ji bang,

whereas weigh lexically selects but does not 0-mark how many pounds.. A

78
closer inspection reveals that even this stipulation does not solve the problem.
For one thing, zhong can also be read as 'gain weight', when suffixed by the
perfective aspect -le, as exemplified below:

(6) a. Akiu zhong-le.


Akiu weigh-Prf
Akiu has gained some weight.
b. Akiu zhong-le liang bang
Akiu weigh-Prf two pound
Akiu has gained two pounds.

Here the amount phrase liang bang 'two pounds' is optional, and hence neither

lexically selected nor 0-marked by zhong-le. Nonetheless, there is still no

locality effect what-so-ever when wh-island constructiins are involved, as in (7):

(7) ni xiang-zhidao [shei zhong-le duoshao/ji bang] ne?


you want-know who gain-Prf how-many pound Qwv
a. Who is the person x such that you wonder how many pounds x has
gained?
b. What is the number/amount x such that you wonder who has gained
x pounds?

The parallel between zhong 'weight' and zhong-le 'gain weight' again
emerges in the long-distance question construals of (8-11), where Complex NP
islands are violated, but still no deviance is detected:

79
(8) [DP [cP zhong duoshao/ji bang] de zhu] cai keyi canjia bisai ne?
weigh how-many pound PNM pig just can join competition Qm
What is the number/amount x such that pigs which weigh x pounds just
can join the competition?

(9) [DP[CP zhong-le duoshaoi]i bang] de zhu] cai keyi canjia bisai ne?
gain-Prf how-many pound PNM pig just can join competition 0Q
What is the number/amount x such that pigs which have gained x pounds
just can join the competition?

(10) [DP [CP Akiu zhong duoshao/ji bang] de shuofa] bijiao kexin ne?
Akiu weigh how-many pound PNM story more believable Qm
What is the number/amount x such that the story that Akiu weighs x
pounds is more believable?

(11) [DP [CP Akiu zhong-le duoshao/ji bang] de shuofa] bijiao kexin ne?
Akiu gain-Prf how-many pound PNM story more believable Qw
What is the number/amount x such that the story that Akiu has gained
x pounds is more believable?

Similarly, English multiple wh-questions also allow long-distance construals of


amount wh's-in-situ in Complex NP constructions, as evidenced by the
following contrast (data due to Howard Lasnik):

(12) a. *How many pounds does who remember [DP a man [cP who weighed t ]]?
b. Who remembers (Dp a man [cP who weighed how many pounds]]?

We may therefore rule out (-marking as a factor in deriving the absence of

LF locality effects. In fact, there is substantial evidence indicating that 0-marking

does not play a role in licensing Chain-formation either: As Rizzi (1990:77)

30
observes, there is an unexpected asymmetry between French wh-adverbials
avec qui 'with whom' and comment 'how'. First note that adveros like bien 'well'

are lexically selected (or 0-, narked) by se comporter 'behave', while PPs ilke

avec les amis 'with friends' are not, as evidenced by (13):

(13) Jean se comporte *(bien) (avec les amis).


Jean behaves well with friends.

However, the locality effects displayed by their corresponding wh-forms are

totally unexpected given the 0-government requirement:

(14) a. ?Avec quij ne sais-tu pas [commenti [PRO te comporter ti tj ]?


With whom don't you know how to behave?
b. *Comment- ne sais-tu pas [avec quij [PRO te comporter ti tj ]?
How don't you know with whom to behave?

As shown by the contrast between (14a,b), the adverb comment, though


lexically selected, undergoes successive cyclic movement, and observes both
the ECP and Subjacency accordingly. In contrast, the extraction of the adjunct
PP avec qui counts as long wh-movement due to its referentiality (or
individuality in Heim-Frampton's sense). And hence the minor Subjacency
violation in (14a). We may thus summarize the whole pattern of contrasts ir, the
following table:

81
(15) Successive cyclic wh-mvt Long wh-mvt Unselective binding
Subjacency/CED yes yes no
ECP/R. Minimality yes no no

The noun-adverb asymmetry, on the other hand, plays an important role in


construing amount wh's: First compare the following examples to (3b) and (5b)
(Chinese and English data due to Jim Huang and Noam Chomsky
respectively):

(16) ni xiang-zhidao [shei (you) duo zhong] ne?


you want-know who have how heavy OQ
a. Who is the person x such that you wonder how heavy x is?
b. #What is the degree x such that you wonder who is x heavy?

(17) a. *How much does who remember [whether John weighed t ]?


b. *Who remembers [whether John weighed how much]?

As shown by (16), zhong can also be construed as a stative adjective when


modified by adverbs such as duo 'how' and hen 'very' (as in Akiu hen zhong
'Akiu is very heavy'). Although the subject reading (16) is as good as (3a), the
wide scope construal of the wh-adverb duo is in fact blocked, as evidenced by
(16b). This is exactly what we would expect from the noun-adverb distinction
explored in chapter 1. That is, only wh-nominals introduce variables in situ,
whereas wh-adverbs must move to create operator-variable pairs. Moreover,

the parallel between (17a,b), in contrast to the asymmetry between (5a,b), also
lends cross-linguistic support to our position: Only nominals such as how many
pounds, but not adjectivals such as how much, are subject to long-distance in-

82
situ construals. 2
A descriptive generalization is thus in order: The availability of long
movement is determined by the referentiality or individuality of the wh-phrases
involved, whereas the availability of unselective binding hinges upon their
nominality, as illustrated in the following table:

(18) Successive cyclic wh-mvt Long wh-mvt Unselective binding


wh-adverb yes no no
non-referential wh-DP/PP yes no yes
referential wh-DP/PP yes yes yes

On technical grounds, there are essentially two ways to look at the wide
scope construals of amount wh's along the unselective binding approach. One
is to treat duoshao/ji bang as objectual/individual variables (vs. amount/degree
variables), in that numbers count as formal objects. As pointed out by Noam
Chomsky and Jim Huang (p.c.), the matrix Q-operator in (3b) may well quantify
over a set of numbers, as shown by the following derivation:

2 Here we may treat stative weigh as a middle verb in that its subject appears t,
originate from a small clause, as in (i), or derive from a logical object predical J
by an irfinitive, as in ({i):
(i) Johni weighs [sc ti how many pounds/how mnuch].
(ii) Johni weighs ti [PRO how many pounds/how much].
In either case, how many pounds count as a predicative nominal, and how
much a predicative adjectival.
83
(19) [cP Opk [C Qk ni xiang-zhidao [CP Opj [cG Qj [lP shei(j) zhong
[IP
you want-know who weigh
duoshao/ji bang(k) ]]]]]]?
how-many pound

[cP- Ic RCQ[IP ni xiang-zhidao [CP [C' Qj [IP shei(j) zhong


you want-know who weigh
d'joshao/ji bang(k)]]] 'I?
how-many pound

Here we may assume either thai a Q-operator transfers its binding relation to a
Q-Comp through operator feature checking (i.e., Spec-head agreement), or that
the Q-Comp may enter into the binding relation by itself after feature checking.
The crucial point here is that, once the [Q] feature is "checked off" in the CP
Spec, the Q-operator is free to delete, since its or;:, lexical content has been
nullified and there is no concern with the recoverability condition. It is thus
possible to avoi J the scenario where a variable is bound by two operators (i.e.,
the Q-operator end the Q-comp). 3
Given the copy theory developed in Chomsky (1992), the sam,.- analysis
carries over to Er jlish multiple /h-questions, as shown by the following
derivation of (5b):

3 Also note that the distinction between Q-operators and Q-Comps is a technical
necessity if we take the view that se'ectional restrictions can be satisfied only
through Spec-head agreement. If we assume instead that the presence of Q-
Comps in itself fulfills the selectional restrictions, then the above issue does not
arise, since no feature checking is involved in licensing question formation (also
cf. Cheng 1991) As a result, there is no need for postulating Q-operators in
Chinese.
84
(20) LF: fcP whoi [C'Qi [iP whoi remembers [CP whetherj [c' Qj [iP John
weighed how many pounds ]]]]]]?

[CP[C 'Qi,k [IP who(i) remembers [cP ['Qyes-no [iP John weighed
how many pounds(k) ]]]]]]?

First we will put aside the exact treatment of whether in the intermediate Spec of
CP.It suffices to note that whether ranges over two opposite cases or situations,
inducing an indirect yes-no question. We then apply upward deletion to the
subject Chain of the matrix clause, as the [Q] feature on the head who is
checked off. This move also prevents the copy who from being bound by two
operators. The matrix Q-Comp thus licensed further serves as a binder and the
only binder of the amount wh-in-situ, because the intermediate Q-Comp
(expressing a yes-no question) is incompatible with the wh-question construal
of how many pounds.
The other alternative is to analyze the wh-in-situ as a functional variable
(or a D variable in Chomsky's (1992) terms; see also Engdahl 1980, Reinhart
1992,1993), as illustrated below:

(21) [Q [IP ni xiang-zhidao [ce OpF


[cP OpG tQ[C' 0 iPF(shei) zhong
[IC'
[cQ
you want-know who weigh
G(duoshao/ji bang) ]]]]]]?
how-many pound
' [CP [c'QG [IP ni xiang-zhidao [( [c'Q F [IP F(shei) zhong
you want-know who weigh
G(duoshao/ji bang)]]]]]]?
how-many pound

85
Under this view, the Q-Comp in question ranges over a set of functions which
apply to the noun head (cf. chapter 1, section 3), inducing an amount or degree
reading. The English multiple wh-question can be treated in a similar way:

(22) LF: [cP whoi [c' Qi [IP whoi remembers [cP whetherj [c' Qj [IP John
weighed how many pounds ]]]]]]?

-[CF [C'Qi, [IP who(i) remembers [cP Ic'Qyes-no [IP John weighed
G(how many pounds) ]]]]]]?
or - [CP [C'QF,G [IP F(who) remembers [cP Ic'Qyes-no liP John weighed
G(how many pounds) ]]]]]]?

Because, as far as duoshao/ji bang and how many pounds are concerned, the
difference between the number-objectual reading and the amount-functional
reading is too subtle to justify an argument in favor of either of the two
approaches, we will leave the choice open here.
A natural prediction along our line is that Chinese measure wh-phrases
should be able to undergo indefinite construals, which is again borne out. First
compare (8a,b) and (11 a,b) with (23a,b) and (24a,b) respectively. The two sets
of sentences are near minimal pairs: In the former, we have an optional wh-
question marker ne and an adverb like cai'just' or bijiao 'more' in the matrix
clauses; in the latter, we have the universal binder dou instead, ranging over
either numbers or amounts (see above):

86
(23) a. zhong duoshao/ji bang] de zhu] dou keyi canjia bisai.
[DP [CP
weigh how-many pound PNM pig all can join competition.
For every number/amount x, pigs which weigh x pounds can join the
competiticn.
b. [DP [CP zhong-le duoshao/ji bang] de zhu] dou keyi canjia bisai.
gain-Prf how-many pound PNM pig all can join competition.
For every number/amount x, pigs which have gained x pounds can join
the competition.
(24) a. [DP [CP Akiu zhongduoshao/ji bang] de shuofa] dou bu kexin.
Akiu weigh how-many pound PNM story all not believable
For every number/amount x, the story that Akiu weighs x pounds is
not believable.
b. [DP [CP Akiu zhong-le duoshao/ji bang] de shuofa]dou bu kexin.
Akiu weigh-Prf how-many pound PNM story all not believable
For every number/amount x, the story that Akiu has gained x pounds
is not believable.

As expected, the noun-adverb asymmetry also show up in dou-quantification:


When we substitute (you) duo zhong '(have) how heavy' for zhong(-le)
duoshao/ji bang, the sentences degrade considerably, as evidenced by the
deviance of (25) and (26):4

4Note that (25) is relatively acceptable without you 'have', as exemplified


below:
(i) ?[[duo zhong] de zhu] dou keyi canjia bisai.
how heavy PNM pig all car, join competition.
For every degree x, pigs which are x heavy can join the competition.
This is because dou zhong alone can be analyzed as a attributive adjective,
which is subject to both long-distance question construals and dou-
quantification:

87
(25) * [DP [CP you duo zhong] de zhu] dou keyi canjia bisai.
have how heavy PNM pig all can join competition.
For every degree x, pigs which are x heavy can join the competition.

(26) * [DP [CP Akiu (you) duo zhong] de shuofa] dou bu kexin.
Akiu have how weigh PNM story all not believable
For every degree x, the story that Akiu is x heavy is not believable.

A precaution here is that cases like (25) and (26) should not be confused
with sentential subject constructions such as (27a) or bare conditionals such as
(27b), which in general allow predicative duo zhong, as exemplified below:

(27) a. [cP Akiu (you) duo zhong] dou wu-quan-jinyao.


Akiu have how heavy all not-concern-matters
It doesn't matter [how heavy Akiu is].
b. Akiu (you) duo zhong, wo jiu (you) duo zhong.
Akiu have how heavy I then have how heavy
However heavy Akiu is, I am (exactly) that heavy.

As pointed out by Jim Huang (p.c.), what undergoes dou-quantification in (27a)


is not duo zhong 'how heavy', but the whole sentential subject. In other words,

(ii) [[Akiu yang-le [(*you) duo zhong de zhu]] de shuofa] bijiao kexin (ne)?
Akiu breed-Prf have how heavy PNM pig PNM story more believable Qwh
What is the degree x such that [the story [that Akiu has bred x heavy a pig]]
is not believable?
(iii)Akiu [(*you) duo pianyi de chezi] dou bu mai.
Akiu have how cheap PNM car all not buy
For every degree x, Akiu do not buy x cheap a car.
The presence of you in (25) thus guarantees that duo zhong is a predicative
adjective rather than a attributive one. As for the cause of this asymmetry, we do
not have a comprehensive answer except the hunch that attributive adjectivals
tend to be nominal, while predicative adjectivals tend to be verbal.
88
dou ranges over questions (or more precisely, answers to the questions) in
(27a). Similarly for the bare conditional, it is the antecedent and consequent
clauses which undergo universal quantification in (27b), not the predicates
which they contain (also cf. chapter 1, section 2.3): A necessity operator
quantifies over two sets of questions in a tripartite style, each of which
corresponds to one clause of (27b):

(28) For every x, if x is an answer to how heavy Akiu is, then x is an answer
to how heavy I am.

Our position is further supported by the fact that wh-elements which usually
resist indefinite construals also occur in the same environments. For instance,
A-not-A questions and wh-adverbs like weishenme 'why' as a rule resist dou-
quantification:

(29) * [DP [CP Akiu lai-bu-lai] de shuofa] dou bu kexin.


Akiu ccme-not-come PNM story all not believable
Both the story that Akiu comes and the story that Akiu does not come are
not believable.

(30) * [DP [c weishenme zhong-le] de zhu] dou keyi cani;a bisai.


why gain-weight-Prf PNM pig all can join competition.
For every reason x, [pigs [which have gained weight for x]] can join the
competition.

However, just as duo zhong 'how heavy' in (27a,b), they are allowed in bare
conditionals and in sentential subjects under the scope of dou, as illustrated
below:

89
(31) a. ccp Akiu lai-bu-lai] dou wu-quan-jinyao.
Akiu come-not-come all not-concern-matters
It doesn't matter [whether Akiu comes or not].
b. Akiu weishenme gaoxing, wo jiu weishenine shang-xin.
Akiu why happy I then why hurt-feeling
I am hurt for whatever reasons Akiu is pleased.

Cases like (27a,b) and (31a,b), therefore, not only do not count as
counterexamples of the noun-adverb asymmetry, but also shed new light on the
intriguing nature of dou-quantification and bare conditionals.
Finally, there are a few mysteries which deserve mention. First of all, for
some reason A-not-A questions are incompatible with bare conditionals, as
shown below:

(32) * Akiu qu-bu-qu, wo jiu au-bu-qu.


Akiu go-not-go I then go-not-go

(32), if grammatical, would mean 'IfAkiu goes, I will go; if Akiu does not go, I will
not go'. This suggests that there is still some fundamental difference between
yes-no questions and wh-questions in conditional construals. Second, duo-
shac, literally translated as 'many-few', differs from ji in resisting indefinite
construals in existential contexts:

90
(33) a. Akiu shi-bu-shi zhong-le ji bang?
Akiu is-not-is gain-Prf a few pound
Isn't it the case that Akiu has gained a few pounds?
b. ruguo Akiu zhong-le ji bang,.....
if Akiu gain-Prf a few pound
If Akiu has gained a few pounds, then .
(34) a. *Akiu shi-bu-shi zhong-le duo-shao bang?
Akiu is-not-is gain-Prf many-:'ew pound
Isn't it the case that Akiu has gained a few pounds?
b. *ruguo Akiu zhong-le duo-shao bang,....
if Akiu gain-Prf many-few pound
!fAkiu has gained a few pounds, then .....

Since A-not-A questions are incompatible with w-questions, ji bang can on!y
be licensed by existential closure, and hence the indefinite construal as 'a few
pounds' in (33a), Duo-shao bang, in contrast, is denied this option.
Consequently, both interrogative and indefinite construals are blocked, as
shown by the deviance of (34a). Similarly, the most prominent reading of ji bang
embedded within a ruguo-conditional is indefinite, as shown by (33b).5 The
same construal is not available for duoshao bang, as evidenced by (34b).
This asymmetry should be further compared with cases where no triggers
like A-not-A questions and ruguo are involved:

5 For some speakers marginal interrogative construals are allowed in ruguo-


conditionals. Namely, (33b) can be read as a matrix question for them:
(i) What is the number/amount x such that if Akiu has gained x pounds, then

91
(35) Aku zuotian mai-le ji-ben shu (ne)
Akiu yesterday buy-Prf a few-CL book Qh

a. Akiu bought a few books yesterday.


b. How many books did Akiu buy yesterday?

As shown above, ji can also be construed as a genuine indefinite (vs. polarity


items), in alternation with its question construal. Here again duo-shao is denied
the indefinite option, and only the interrogative reading survives, as illustrated
by the contrast between (36a,b):

(36) Akiu zuotian mai-le duo-shao-ben shu (ne)


Akiu yesterday buy-Prf many-few-CL book Qwh

a. #Akiu bought a few books yesterday.


b. How many books did Akiu buy yesterday?

As we can tell from the English translation, duo-shao 'many-few' is really a


disjunctive compound, which migh; account for its remote resemblance to the A-
not-A question in (32). This intrinsic property is most prominent when duo-shao
is used as a sentential adverbial, meaning 'more-or-less':

(37) Akiu duo-shao zuo-le yixie hao shi.


Akiu many-few do-Prf some good thing
Akiu more or less has done some good deeds.

Since this adverbial usage is the only case where duo-shao can be said to be
existential (roughly an equivalent of 'in some cases'), it may well be the case
that existential duo-shao has been "specialized" into its present adverbial form,

92
while existential ji is "generalized" to cover even genuine indefinite construals.
In sum, we have shown that we do have a case where neither the lack of

referentiality nor the absence of 0 -marking makes any difference as to the

licensing of long-distance A'-dependencies. The implication is significant: There


is an entire different breed of long-distance construals, namely, unselective
binding, which cannot be leveled with long wh-movement. We will continue to
examine some suggestive facts in the following sections.

93
2. Interrogative Construals in Dou- and Mo-Quantification

As we assimilate interrogative construals of wh's-in-situ to typical cases of


unselective binding like dou-quantification, we also commit ourselves to the
prediction that the interaction between both should not display the ECP and/or
Relativized Minimality effects. This prediction appears to be false at first glance:

(38) [cpshei zai shafa-shang shui] dou ke-yi (*ne)


who at sofa-on sleep all will-do Qh
a. Whoever sleeps on the sofa, it will do.
,,eeps on the sofa?
b. #Who is the person x such that it will do if
A

As shown above the wide-scope question construal (38b) appears to be


blocked by dou. A similar observation is also made by Nishigauchi (1990:148)
concerning mo-quantification in Japanese, as exemplified below:

(39) kimi-wa [cpdare-ge ki-te]-mo ik-a-nai no?


you-Top who-Nom come-all go-not Q
a. Are you not going, whoever may come?
b. #For which x, x a person, are you not going if x is coming?

First note that unlike the necessity operator which licenses bare
conditionals, dou and mo only take scope over the embedded CPs of (38) and
(39) respectively. As a result, when shei 'who' and dare 'who' undergo long-
distance construals, dou and mo becomes vacuous quantifiers. Besides,
according to the "answer variable" hypothesis presented in the last section, the
embedded CPs are actually indirect questions. The wide scope question
construals (38b) and (39b) thus fail to satisfy the selectional restrictions on the

94
embedded CPs, resulting in the -ollowing nonexistent readings:

(40) # Who is the person x such that if for every y, y satisfies the property of
[x sleeps on the sofa], it will do?
(41) # Who is the person x such that if for every y, y satisfies the property of
[x is comingi, you are not going?

Either way, (38b) and (39b) are correctly ruled out. Along this line, a natural
question is why question construals are ever allowed in bare conditionals such
as (42b"):

(42) shei xian lai, shei jiu keyi xian chi (ne)
who first come who then can first eat Qm
a. Whoever comes first, then (s)he is allowed to eat first.
b. Who is the person x such that if x comes first, x is allowed to eat first?

The soltution to this puzzle is where the answer variable analysis really shines:
1 he "a:ross-the-board" behavior of universal quantification in (42a) and that of
the question construa! in (42b) can be both attributed to the fact that it is
answers that matter here. This point is illustrated by (43a,b), which correspond
to (42a,b) respectively (situatihn variables omitted):

(43) a. For every x, if x is an answer to the question who comes first, then x is
the answer to the question who is allowed to eat first.
b. What is x such that if x is an answer to the question who comes first,
then x is the answer to the question who is allowed to eat first?

The above move in turn yields another prediction: if there is more than one

95
wh-in-situ in dou- and mo-clauses (i.e., antecedent clauses), the wide scope
construals in question should be possible, which is indeed the case in Chinese:

(44) [cP (tamen) shei zai nali shui] dou ke-yi (ne)
they who at where sleep all will-do Cvh

a. No matter (among them) who sleep where, it will do. (If for every x,
x is an answer to the question who sleeps where, it will do.)
b. Who is the person x such that it will do wherever x sleeps?
c. Where is the place y such that it will do whoever sleeps at y?

As shown above, in addition to the paired indirect question reading (44a), each
of the wh's-in-situ may assume the wide scope, as in (44b) and (44c). Also as
expected, (44) does not allow a paired matrix question like (45), for reasons just
mentioned:

(45) * Who is the person x and where is the place y such that if for every y,
y is an answer to the question [x sleeps at y], it will do?

We can further eliminate the (a) clause reading by embedding (44) as a an


indirect question, as shown by (46);

(46) Akiu xiang-zhidao R[cP shei zai na!i shui] dou ke-yi].
Akiu want-know who at where sleep all will-do
a. Akiu wonders who is the person x such that it will do wherever x sleep.
b. Akiu wonders where is the place y such that it will do whoever sleeps
aty.

Here the long-distance question construals (46a) and (46b) stand out as the
only two readings available.

96
As for Japanese, the prediction is only partially borne out, as exemplified
beiow (data due to Masa Koizumi):

(47) [cP dare-ga nani-o tabe-te! ..o ii (no)


who-Nom what-Acc eat-lnfl]-all good Q
a. No matter who ate what, it will do.
b. Who is the person x such that it will do whatever x eat?
c.#What is the thing y such that it will do whoever eat y?

Just as its Chinese counterpart shai in (44), the subject wh-in-situ dare may
assume the wide scope in (47b), in addition to its narrow scope construal (47a).
In contrast, the object iPni 'what' only allows the narrow scope reading.
Consequently, there seems to be a subject-object asymmetry w0'h respect to
unselective binding. But as Masa Koizumi (p.c.) points out, the (c) clause
reading becomes available when the object is scrambled to the left of the
subject:

(48) [cP nani-o dare-ga tabe-te]-mo ii (no)


what-Acc who-Nom eat-lnfl]-all good Q
a. No matter who ate what, it will do.
b.#Who is the person x such that it nvill do whatever x eat?
c. What is the thing y such that it will do whoever eat y?

This indicates that the asymmetry may have something to do with (anti-
)superiority effects in Japanese: The embedded CP, being an irldirect question,
ce the selectional restriction must
may well involve invisible wh-movement, sli
be fulfilled by feature-checking before SPELL-OUT (recall our assumption that
operator features are strong in Japanese (cf. chapter 1, section 4); see also

97
Watanabe 1991). As a detailed account is beyond the scope of this chapter, we
will drop the issue here.
Another factor which can be isolated in (38) concerns the presence of
indirect questions under the scope of dou. Recall that dou takes scope over not
only CPs, but also DPs, where the noun-adverb asymmetry holds, and
answer/situation variables are irrelevant (cf. section 1). In the latter cases, long-
distance construals should be available even when there is only one wh-in-situ
within the scope of dou, as long as the ban against vacuous quantification is
observed. This is indeed the case, as evidenced by (49):

(49) [DP[CP Dufu zai nali xie] de shi] dou shi yiliude (ne)
Dufu at where write PNM poem all be first-rate Qw
a. For every x, x a place, for every y, y a poem which Dufu wrote at x,
y isfirst-rate.
b. Where is the place x such that for every y, y a poem which Dufu
wrote at x, y is first-rate.

As shown above, dou may either quantify over both zai nali 'at where' and the
head noun (i.e., the bare indefinite shi 'poem'), as in (49a), or quantify over the
head noun alone, as in (49b). It is in the latter option that a direct question is
possible.
Our observation is further consolidated by the fact that even paired
question readings are allowed when there are more than one wh-in-situ in the
same configuration. Take (50) for example, where Dufu is replaced by shei
'who': dou may quantify over shei, zai nali, and the head noun shi in one swipe,

98
as in (50a). Alternatively, either one of the wh's-in-situ may assume the wide
scope, as in (50b,c). And most importantly, a paired direct question is also
possible when dou quantify over the head noun alone, as in (50d):

(50) [DP [cp shei zai nali xie] de shi] dou shi yiliude (ne)
who at where write PNM poem all be first-rate Q,
a. For every x, x a person, for every y, y a place, for every z, z a poem
which x wrote at y, z is first-rate.
b. Who is the person x such that for every y, y a place, for every z,
z a poem which x wrote at y, z is first-rate?
c. Where isthe place y such that for every x, x a person, for every z,
z a poem which x wrote at y, z is first-rate?
d. Who is the person x and where is the place y such that for every z,
z a poem which x wrote at y, z is first-rate?

Again, if we further embed the whole sentence as an indirect question, the "free
relative" reading (50a) is suppressed due to the selectional restriction imposed
by the matrix verb, and the long-distance question construals parallel to (50b-d)
stand out most clearly, as exemplified below:

(51) Akiu xiang-zhidao [CP [DP [CP shei zai nali xie] de shi] dou shi yiliude].
Akiu want-know who at where write PNM poem all be first-rate
a. Akiu wonders who is the person x such that for every y, y a place,
for every z, z a poem which x wrote at y, z isfirst-rate?
b. Akiu wonders where is the place y such that for every x, x a person,
for every z, z a poem which x wrote at y, z isfirst-rate?
c. Akiu wonders who is the person x and where is the place y such that
for every z, z a poem which x wrote at y, z is first-rate?

99
To sum up, we have demonstrated that the interaction between dou-
quantification and wh-question formation displays the same characteristics as
that among wh's-in-situ themselves. Namely, neither Subjacency or the
ECP/relativized minimality effects are detected. Although the situation is less
clear on the part of Japanese mo-quantification, it seems safe to attribute the
difference to the strong status of its operator features.

100
3. A'-Bound Pro or Pure Variable?

Given our observation so far, one might sketch a working hypothesis


based on Cinque's (1990) analysis of parasitic gaps and tough-constructions:
That is, Chinese allows an extensive A'-bound (resumptive) pro strategy, not in
syntax, but in LF. Along this line, all wh's-in-situ undergo LF movement in
Chinese. Since only wh-nominals have corresponding proforms, the resumptive
pro strategy is not available for wh-adverbs. And hence the noun-adverb
asymmetry. The proposal, though stipulative in nature, appears to be technically
sound. Therefore, we should look further into other factors to determine whether
it is consistent with the general properties of Chinese wh-construals.
On conceptual grounds, an immediate concern is that we are to lose the
generalization that Chinese interrogative and indefinite wh-construals are two
instances of one type of A'-dependency, i.e., unselective binding (or simply
binding in the context of the Lexical Courtesy Hypothesis). In other words, we
have to stipulate two classes of wh-nominals: One with intrinsic quantificational
force, and the other without. Moreover, Subjacency again has to be ordered
before SPELL-OUT, which in turn renders our account of Hindi wh-questions
irrelevant. Most importantly, as Huang (1984) observes, Chinese allows an
empty pronominal only to the extent that it is associated with the closest
potential controller:

(52) * zhe-ge renk, Akiu xihuan [DP hao-ji-ben [cP ek xie elj] de shuj].
this-CL person Akiu like quite-a few-CL write PNM book
This person, Akiu likes quite a few books that (he/she) wrote.

101
As shown above, the control relation between the topic and the embedded
empty subject is blocked by the matrix subject Akiu, a potential controller for the
subject EC. When we shift the complex NP to the subject position, the deviance
disappears, as evidenced by (53):

(53) zhe-ge renk, [DP hao-ji-ben [cP ek xie ej] de shuj] dou de-le jiang.
this-CL person quite-a few-CL write PNM book all win-Prf award
This person, quite a few books that (he/she) wrote has won awards.

This is because the original blocking factor is eliminated. The only exception to
this minimality requirement (the Generalized Control Ruie (GCR) in Huang's
terms) occurs when the controller is inanimate or inhuman. This point can made
clear by comparing (52) with (54), where we have zhe-ben shu 'this book'
instead of zhe-ge ren 'this person' as the topic:

(54) zhe-ben shUk, Akiu renshi [DP hao-ji-ge [cP ej xihuan ek ] de renj].
this-CL book Akiu know quite-a few-CL like PNM person
This book, Akiu knows quite a few people who like (it).

In other words, a resumptive pro, which behaves more like a lexical pronoun
and does not observe the GCR, can only be associated with an
inanimate/inhuman antecedent in Chinese. However, there is no such
restriction in construing Chinese wh's-in-situ, as exemplified below:

102
(55) Akiu xihuan [DP [cP shei xie ej] de shuj] ne?
Akiu like who write PNM book Qw
Who is the person x such that Akiu likes books that (he/she) wrote?

On empirical grounds, first we would like to point out that the control relation, as
well as (lexical) pronominal binding, does not display specificity effects:

(56) zhe-ge renk, [DP naxie [cP tak/ek xie ej] de shuj] dou de-le jiang.
this-CL person those (s)he write PNM book all win-Prf award
This person, those books that (s)he wrote has won awards.

(57) zhe-ben shUk, Akiu renshi [DP naxie [cP ej xihuan ek ] de renj].
this-CL book Akiu know those like PNM person
This book, Akiu knows those people who like (it).

As shown above, the presence of an intervening demonstrative such as naxie


'those' does not affect long-distance control and pronominal binding. In
contrast, long-distance wh-construals out of complex NPs are in general
blocked by demonstratives:

(58) * [DP naxie [cP shei xie ej] de shuj] dou de-le jiang ne?
those who write PNM book all win-Prf award Q
Who is the person x such that those books that x wrote has won awards?
(59) * Akiu xihuan [Dp naxie [cP shei xie ej] de shuj] ne?
Akiu like those who write PNM book Qk,@
Who is the person x such that Akiu likes those books that x wrote?

This contrast thus lends strong support to the distinction between pronominal
variables and so-called "pure" (i.e., [-pronominal]) variables in Cinque's sense.

103
Another piece of evidence comes from double island constructions. As
Cinque (1990) points out, A'-bound pro construals are not entire island-free:
They are capable of violating one complex NP island, as evidenced by the
following Italian examples (see also Longobardi 1983, Kayne 1984):

(60) ?Carlo, chek abbiamo ricoverato tk [con la speranza di


Carlo who we hospitalized with the hope of
poter salvare ek], ...
being able to save

(61) ?Co rio, chek abbiamo discusso pii vrc'4 [Ia 9ossibilith
Carlo who we discussed several times the possibility
di ammettere ek nel nostro club],...
of admitting to our club

But when there are more than one island involved, the construals in question
are blocked, as illustrated below:

(62) * Carlo, chek abbiamo ricoverato tk [con la speranzadi aumentare


Carlo who we hospitalized with the hope of increasing
[le probabilith di salvare ek]], ...
theprobabilities of saving
(63)* Carlo, chek ci siamo presentati [con la speranzadi aumentare
Carlo who we turned up with the hope of increasing
[le probabilitA di salvare ek]], ...
the probabilities of saving

Cinque proposes that the object pro may either undergo wh-movement or pied-

104
piping in LF. and it is the latter option that evades strong island effects in (60)
and (61) by moving the entire complex NP. On the other hand, even LF pied-
piping cannot escape from double islands such as those in (62) and (63), since
the higher complex NP blocks its way. The samne observation, in contrast, does
not obtains for Chinese wh's-in-situ, as evidenced by the well-formedness of the
following double island constructions:

(64) [DP [CP ek xihuan [DP [CP shei xie ei] de shui]] de renk
like who write PNM book PNM person
zui rongyi jiao-dao pengyou?
most easily make-reach friend
Who is the person x such that [people [who like [books [x wrote]]]]
make friends most easily?
(65) ni bu xiangxin [DP [CP [DP [CP shei xie ei] de shui] zui
you not believe who write PNM book most
chang-xiao] de shuofa]?
well-selling PNM story
Who is the person x such that you do not believe [ the story [that
[the book [which x wrote]] is best-selling]]?

This again confirms our view that neither LF movement nor pied-piping involves
in construing Chinese wh-nominals, which in turn casts doubt upon any effort to
reduce unselective binding to A'-bound pro binding.

Even more interestingly, Japanese wh-questions, which are often cited as


typical instances of large-scale LF pied-piping (see, for example, Nishigauchi
1986,1990), pattern with their Chinese counterparts instead of their Italian

105
counterparts, in not displaying double strong island effects, as exemplified
below (data due to Masa Koizumi):

(66) [DP [CP [DP [CP dare-gakaita] hon- o sukina] hito-ga tomodati-o
who-Nom wrote book-Acc like person-Nom friend-Acc
iti-ban tukuri yasui no?
most make easy Q
Who is the person x such that [people [who likes [books [x wrote]]]]
makes friends most easily?

(67) [DP [cP [DP [CP dare-ga kaita] hon-ga] itiban-yoku urete-iru toyuu]
who-Nom wrote book-Nom most-good sell-be that
uwasa-o] sinzite-i-nai no?
rumor-Acc believe-be-not Q
Who is the person x such that you do not believe [the rumor [that
[the book [which x wrote]] is best-selling]]?

Consequently, Watanabe's (1991) invisible wh-movement hypothesis does


have advantage over the traditional pied-piping approach at least in one
respect: It correctly predicts the well-formedn6ss of (66) and (67). This is
because the Q-operators responsible for the long-distance construals can be
generated in the upmost DP Spec, and undergo subsequent cyclic movement
to the matrix CP Spec without violating either one of the complex NP islands (cf.
chaper 1, section 2.5). On the other hand, Nishigauchi's feature-percolation
hypothesis makes the same prediction as Cinque's analysis, which is based on
the notion of g-projection in Kayne and Longobardi's sense. As a result, (66)
and (67) are wrongly ruled out under the LF pied-piping approach.

106
4. Wh-Extraciton from Drived Nominals

Another difference between chain formation and unselective binding lies in


wh-extraciton from derived nominals. As Stowell (1989) observes, the following
contrast can be explained if we assume that the DP Spec in (68b) is filled by the
subject of the picture-NP, i.e., Mary, while it is left empty in (68a) before
extraction applies:

(68) a. WhOk did you sell [DP Vk [D'a [N' picture of tk fl?
b. *Whok did you sell [DP Mary [DO 'S[N' picture of tk ]]?

Consequently, the object uf the picture-NP can extract freely in (68a) through
the "escape hatch", i.e., the DP Spec. On the other hand, since the escape
hatch is not available in (68b), subsequer,' extraction is blocked by the ECP and
Subjacency.
The above contrast does not seem to hold in Chinese long-distance
question construals as evidenced by (69a,b):

(69) a. ta caina-le [DP yi-xiang [D'[pp dui shei] de piping ]?


(s)he adopt-Prf one-CL about whom PNM criticism
WhOk did (s)he adopt a criticism of t?

b. ta caina-le [DP Akiu (de) [D' [pp dui shei] de piping ]]?
(s)he adopt-Prf Akiu Poss about whom PNM criticism
*Whok did s/he adopt Akiu's criticism of tk?

Here the syntactic configuration of (69b) is almost identical to that of (69b),


except that the PP complement occurs to the left of the head noun. However, the

107
long-distance construal of shei 'who' does not display any ECP/Subjacency
effect.
In addition, there is a further contrast between the English example (70a)
and its Chinese counterpart (70b), where the DPs in question are headed by
so-called strong determiners:

(70) a. *Whok did you sell [DP every/several/all/most [D' [N' picture(s) of tk ]]]?

b. ta caina-le [DP mei-yi-xiang/hao-ji-xiang/suoyou/daduoshu


(s)he adopt-Prf every-one-CLlquite-a few-CL/all/most

[D' [PP dui shei] de piping]]?


about whom PNM criticism
*Whok did (s)he adopt every/quite a few/all/most criticism(s) of t?

By assuming that a strong determiner either occupies the Spec itself or does not
project a Spec position, we correctly rule out (70a). The same observation,
however, does not apply to (70b).
Note that the absence of locality effects in (69b) and (70b) cannot be totally
subdued by claiming that the CED/Subjacency does not hold in LF. Let's
consider first the contrast between (71a,b):

(71) a. Akiu [pp dui shei] bu manyi?


Akiu about who not satisfied
Who is the person x such that Akiu is not satisfied with x?
b. *sheik, Akiu [pp dui tk] bu manyi?
who Akiu about not satisfied

108
As shown by (71b), overt wh-fronting over the PP node results in strong
deviance. In comparison, the deviance caused by extraction out of adjunct
clauses is relatively minor:

yi jian tk ]] iu bu zou-le?
(72) a. ?sheik , Akiu [cp Opi [iP ti
who Akiu once meet then not leave-inc
Who is the person x such that Akiu didn't leave once he met x?
b. ?sheik , Akiu [cP Opi [IP ti jian-ye-mei-jian tk ]] jiu zou-le?
who Akiu meet-also-not-meet then leave-Inc
Who is the person x such that Akiu left without meeting x?

This indicates that the ECP is at least partially responsible for the strong

deviance of (71b): On the one hand, dui 'about' does not count as a 0-governor;

on the other, antecedent government is blocked by the PP node. In contrast,


since shei islicensed by the verb jian 'meet' in (72a,b), the extraction in
question only crosses one barrier (i.e., the CP node), inducing the
CED/Subjacency effect.
As a result, if (69b) and (70b) involves LF wh-extraction, we would detect
deviance at least stronger than that of (72a,b) due to an ECP violation. This
prediction, as we have seen above, is not borne out.

109
5. Reflection on a Syntax-LF Asymmetry

As far as wh-construals are concerned, there is a recent trend in the


literature to reduce argumentality to referentiality (see, among others, Aoun
1986, Cinque 1989, Rizzi 1990, Tsai 1994), or nominality to pronominality (cf.
Cinque 1990). These proposals, though extremely useful in accounting for a
wide range of data, do not seem to fully capture a curious generalization
observed in Huang (1982): That is, when and where pattern with why and how
in syntax, while they pattern with who and what in LF. Here the paradox is that if
when and where are referential (or leave [+pronominal] variables to a similar
effect), they should pattern with arguments throughout ,drivaiion, either in
undergoing long movement, or in binding resumptive pro's. If not, they should
only be subject to cyclic movement, patterning with adjuncts. Therefore, other
factors have to be introduced to derive this syntax-LF asymmetry.
There are essentially two ways to approach the problem. One is based on
Huang's (1982) intuition that when and where behave like PPs in syntax, but
like NPs in LF. In other words, they are syntactically PPs, but semantically NPs.
Huang proposes that this is because when and where are NPs with empty
prepositions, as in [pp [p e] wh ].Consequently, they may extract freely from the
"bare" PP node in LF, given that the CED/Subjacency holds only in syntax. The
other is to follow WHAL's (1987) split ECP approach, attributing the adjunct
behavior of when and where to the head-government requirement in PF, and
their argument behavior to the (generalized) binding requirement in LF (see
also Tsai 1994).
Nevertheless, the problem is compounded by the fact that this marginal
behavior of wh-adjuncts is not limited to when and where. As observed by Lin
(1992), there is a how-why asymmetry in construing Chinese wh's-in-situ:

110
Zenmeyang 'how' patterns with arguments, while weishenme 'why' retains its
adjunct characteristics. Tsai (1994) further points out that the how-why
asymmetry is a subcase of a more general asymmetry between instrumental
how and purpose why on the one hand, and manner how and reason why on
the other. The whole pattern of contrasts is summarized in the following tables
(see below for some detailed discussions):

Chinese in-situ wh-construal Strong/Wh-Islands Non-Bridge Verb Others


who ok ok ok
what ok ok ok
where ok uok
when ok ok ok
how many-DP ok ok ok
how-AP (predicative) * ok
resultative how ok ok ok
instrumental how ok ok ok
purpose why ok ok ok
manner how * ok
reason why * * ok
This paradigm should be further compared with overt wh-fronting in Chinese:

Chinese wh-fronting Simple Sentence Bridge Verb Others


who ok ok
what ok ok *
where* * *
when ok with modals ok with modals *
how many-DP * * *
how-AP (predicative) * * *
resultative how ok ok *
instrumental how * * *
purpose why * * *
manner how ok postverbally ok postverbally *
reason why * * *

111
Given what we have said in chapter 1, the above syntax-LF asymmetry
may well reflects a fundamental distinction between Chain-formation and
unselective binding in the context of the LCH. That is, wh-fronting is subject to
Economy considerations such as "minimizing Chain links" in the sense of
Chomsky & Lasnik (1991), while in-situ wh-construals are relatively cost-free.
This can be seen by comparing the far more rigid locality displayed by wh-
fronting with the unbounded in-situ construals of all but a few wh-adverbs.
To reduce ensued complexity, we will single out the most problematic
cases such as zenmeyang 'how' and weishenme 'why'. In the following
discussion, we essentially reproduce the observations made in Tsai
(1993a,1994), and see how the asymmetries can be accommodated in the
current framework.
Let's start with zenmeyang, which is three-way ambiguous. When
construed with a manner reading, it acts like an adjunct, displaying island
effects; when construed with an instrumental reading, it patterns with
arguments, lacking any island effect. This point is illustrated by the contrast
between (73a) and (73b), where a complex NP island is involved. Only PPs like
yong xiao huo 'with low heat' and yong shaguo 'with a sand pot', but not
manner adverbs like xiaoxinyiyi-di 'very carefully', are possible answers to (73).
In contrast, both readings are valid in simple sentences like (74):

(73) ni zui xihuan [NP [CP Opi [1P ta zenmeyang duen ti ]] de niuroui ]?
you most like she how stew PNM beef
a. What is the means x such that you like best [beef [which she stewed
by x]]?
b. #What is the manner x such that you like best [beef [which she stewed
in x]]?

112
(74) ta shang-ci zenmeyang dun niuou?
she last-time how stew beef
a. By what means did she stew beef last time?
b. In what manner did she stew beef last time?

Futhermore, it is not uncommon for zenmeyang to function like a predicate,


either intransitive as in (75a) or transitive as in (75b): In (75a), zenmeyang
questions the current state of the subject; in (75b), it questions the consequence
affecting the object:

(75) a. Lisi zenmeyang le?


Lisi how Inc
What happened to Lisi?

b. nimen neng [pp ba wo ] zenmeyang?


you(pl.) can BA me how
What can you do to me?

This predicate usage is also responsible for the resultative reading of


postverbal zenmeyang in (76a), while the adverbial usage leads to the manner
reading in (76b):

(76) a. ta niuroui duen-de [proi zenmeyang]?


she beef stew-DE how
Till what state did she stew beef?

b. ta niurou duen-de zenmeyang?


she beef stew-DE how
In what manner did she stew beef?

Also note that postverbal zenmeyang is exclusively introduced by the V-de


complex, which, for some reason, suppresses the instrumental reading and
introduces the resultative reading (see below).
113
Interestingly enough, when postverbal zenmeyang is further embedded in
a relative clause, only the resultative reading emerges. That is, it patterns with
instrumental zenmeyang (and hence arguments) rather than manner
zenmeyang, as shown by the contrast between (77a) and (77b):

(77) ni zui xihuan [NP [CP OPi [Ip ta ti duen-de zenmeyang]] de niuroui ]?
you most like she stew-DE how PNM beef
a. What is the state x such that you like best [beef [which she stewed till x]]?
b.#What is the manner x such that you like best [beef [which she stewed
inx]]?

This observation is reflected by the fact that the only possible type of answers to
(77) is an AP like lan-yi-dian de 'a little more mushy', but never a manner
adverb like xixin-yi-dian-di 'a little more attentively'. The same pattern of
contrasts obtains for other instances of strong islands, such as the sentential
subject in (78) and the appositive clause in (79):

(78) [zhe-jian shi, women yao zuo-de zenmeyang] cai ling-ren-manyi?


this-CL matter we need handle-DE how just make-people-satisfied
a. What is the state x such that it isjust satisfying [for us to handle this
matter till x ]?
b.#What is the manner x such that it isjust satisfying [for us to handle this
matter inx ]?

(79) ni bijiao xiangxin [[na-dao cai, tamen zuo-de zenmeyang] de shuofa]?


you more believe that-CL dish they cook-DE how PNM story
a. What is the state x such that you believe more [the story [that they
cooked that dish till x ]]?
b.#What isthe manner x such that ysu believe more [the story [that
they cooked that dish in x ]]?

114
What should be further included in this discussion is overt extraction of
postverbal zenmeyang. Generally speaking, Chinese wh-fronting patterns with
topicalization to the extent that no control (non-movement) construal is allowed.
Namely, it observes Subjacency and the head-government requirement. As we
can tell from (80) and (81), the object shei 'who' may move as long as
Subjacency and the CED are respected; in contrast, preverbal zenmeyang
cannot undergo wh-fronting even in a simple sentence, no matter what reading
it is associated with:

(80) a. sheii, ni zui xihuan t ?


who you most like
Who do you like most?

b. shell, ni kan [tamen zui xihuan ti]?


who you think they most like
Who do you think they like most?

(81) a. *zenmeyangi, tamen yinggai ti zuo zhedao cai?


how they should cook this-CL dish
How should they cook this dish?

b. *zenmeyangi, ni kan [tamen yinggai ti zuo zhedao cai]?


how you think they should cook this-CL dish
How do you think they should cook this dish?

On the other hand, overt extraction of postverbal zenmeyang is allowed, with


both manner and resultative readings valid:

(82) ?zenmeyangi (a), ta niurou dun-de ti ?


how Top she beef stew-DE
a. What is the state x such that she stewed beef till x?
b. What is the manner x such that she stewed beef in x?

115
(83) ?zenmeyangi(a), ni kan [ta niurou dun-de ti I?
how Top you think she beef stew-DE
a. What is the state x such that you think [she stewed beef till x ]?
b. What is the manner x such that you think [she stewed beef in x ]?

It is quite clear that this asymmetry is a matter of extraction sites in the


syntactic projection. An immediate answer comes from Huang (1991), in which
an illuminating picture of Chinese postverbal complementation has been
sketched in the sp2rit of Larson (1988). In the relevant discussion, Huang adopts
McConnel-Ginet's (1982) distinction between Ad-VP and Ad-Verb, and
translates it into a VP-shell-type analysis. Essentially, there are two types of
manner adverbs: the "outer" one is a modifier of a verb phrase, while the "inner"
one is a stative predicate which may form a complex predicate with the verb.
This distinction is reflected by the contrast between the inner Ad-Verb in (84a)
and the outer Ad-VP in (84b):

(84) a. John finished the job quickly/ real fast.


b. John quickly/* fast finished the job.

The distribution of Chinese manner adverbials, as illustrated in (85), fall neatly


under the inner/outer pattern:

(85) a. Zhangsan pao-de hen kuai.


Zhangsan runs fast.
b. Zhangsan hen-kuai-di pao le.
Zhangsan quickly ran away.

Here I will take the null hypothesis that this pattern holds for their
interrogative counterparts as well. A specific view of the distribution of
zenmeyang is thus in order:
116
(86) VP

Subj V'

zenmeyang V'

[v e] VP

Obj V'

V-de zenmeyang
As illustrated by (86), perverbal zenmeyang is protected by the higher VP node
from being governed by any functional head, while postverbal zenmeyang is
always head-governed by the lower V node. This move provides a
straightforward account of the contrast between (81a,b) and (82,83).
The semantics of zenmeyang also follows in a natural way: On the one
hand, since the inner position is restricted to a stative predicate, there is no way
to introduce the instrumental reading for postverbal zenmeyang. Accordingly,
the resultative reading comes by default due to the predicate usage of
zenmeyang. On the other, since preverbal zenmeyang is in a modifier position,
the resultative reading is suppressed and the instrumental reading pops out.
Under the unselective binding approach outlined in chapter 1, the
capability of introducing variables is essential to the survival of long-distance
wh-construals in a given island construction. For one thing, zenmeyang consists
of two morphological units: an adverb zenme 'how' and a noun head yang
'manner/way'. It is thus plausible to assume that instrumental zenmeyang
projects to an DP, as sketched in (87);

117
(87) DP

Det N'
I I
zenme N
I
yang

Manner zenmeyang, on the other hand, appears to be a genuine adverb,


patterning with zenme in regard to locality effects. This sort of category-shifting
is not uncommon among languages, as the distinction between some times and
sometimes suggests in English.
Furthermore, given Huang's (1991,1992) ana., :s of postverbal
complementation (cf. (86)), resuliative zenmeyang may well be treated as a
predicate nominal in a small clause, i.e., the result clause (RC) itself, as
illustrated below:

(88) .... VP

Obji V'

V-de RC

Proi zenmeyang

This way, we are able to relate instrumental and resultative zenmeyang in terms
of their categorical status. Namely, they both count as nominals, and
accordingly introduce variables when interpreted in-situ.
Additional support comes from the contrast between purpose and reason
why in Chinese: wei(-le) shenme'for what' contrasts with weishenme 'why' not
only in their readings (i.e., purpose vs. reason), but also in their behavior with
respect to strong islands, as shown below:

118
(89) a. ni zui xihuan [[wei(-le) shenme gongzuo] de ren]?
you most like for what work PNM people
What is the purpose x such that you like best [people [who work for x]]?

b.*ni zui xihuan [[weishenme gongzuo] de ren]?


you most like why work PNM people
What is the reason x such that you like best [people [who work for x]]?

(90) a. ni bijiao xiangxin [[tamen wei(-le) shenme cizhi] de shuofa]?


you more believe they for what resign PNM story
What is the purpose x such that you believe more [the story [that they
resigned for x]?

b.*ni bijiao xinagxin [[tamen weishenme cizhi] de shuofa]?


you more believe they why resign PNM story
What is the reason x such that you believe more [the story [that they
resigned for x]?

(91) a. [women wei(-le) shenme nianshu] cai you yiyi?


we for what study just have meaning
What is the purpose x such that it is just meaningful [for us to study
for x]?

b.*[women weishenme nianshu] cai you yiyi?


we why study just have meaning
What is the reason x such that it isjust meaningful [for us to study for x]?

(89-91) represent relative, appositive, and subject clause constructions


respectively. As we can tell from the contrast between the (a)- and (b)-clauses,
only purpose wei(-le) shenme gets valid wide-scope readings. Unless we stress
the preposition wei 'for' or separate wei from shenme with a suffix -le,
weishenme as a whole counts as a genuine adverb associated with the reason
reading. A similar case is also found in French. The distinction between pour
quoi 'for what' and pourquoi 'why' is reflected in the writing system: Only the

119
former is granted the in-situ option and interpreted as purposive (cf. Aoun
1986).
As for wh-fronting, both weishenme and wei(-le) shenme behave exactly
like preverbal zenmeyang. Namely, they never extract overtly. The head-
government requirement thus seems to be a promising candidate for blocking
Chinese adjunct fronting in general.
This dichotomy based on the categorical status of wh-expressions enjoys a
further advantage over that based on referentiality, in that people sharing the
same locality judgement of (73), (77-79), and (89-91) do not necessarily agree
on our semantic distinction. In particular, the line between reason why and
purpose why is indeed very thin, and at least some of the native speakers
consulted feel that wei(-le) shenme 'for what' could be either causal or
purposeful. On technical grounds, it is also hard to spell out why purpose why is
more "referential" than reason why.
In comparison, the nominality account outlined in chapter 1 (also cf. Huang
1982, Higginbotham 1983, 1985, Reinhart 1992, 1993, Tsai 1993a, 1994, and
to some extent, Cinque 1990) is relatively uncontroversial. Moreover, since one
of defining characteristics of nominals is their ability to refer (or their capability of
naming in a more traditional term), the somewhat vague intuition about
referentiality is also accounted for.
Finally, In view of the diversity displayed by Chinese wh-adjuncts, it is
hardly surprising to find that there are sporadic idiosyncrasies in languages with
more "agglutinating" wh-morphology, such as the asymmetries between *why-
ever and how-ever in English, and that between *naze-mo 'why-ever' and naze-
ka 'why-some' in Japanese (cf. chapter 1).
As a result, before we can conduct a comprehensive research on the origin
of these wh-expressions, it suffices to note that why in *why-ever behaves more
120
like Chinese reason why, while how in how-evei behaves more like Chinese
instrumental and resultative how. Along the same line, naze in *naze-mo
patterns with Chinese reason why, whereas naze in naze-ka patterns with
Chinese purpose why (or more accurately, for what).

121
6. Aftermath

In this chapter, we continue to explore the consequences of the LCH


largely on empirical grounds. We have pointed out the inadequacy of
characterizing long-distance in-situ construals as some kind of long movement
in LF. While they do share some properties such as the lack of ECP effects, they
differs inthe relevance of Subjacency/CED, and in the possibility of construing
amount wh's. To strengthen our position, we further demonstrate that Chinese
wh-questions and dou-quantification do not interact to display any minimal
binding effects (Aoun & Li 1989) or relativized minimality effects (Rizzi 1990), as
long as no vacuous quantification is involved.
It is also argued that the A'-bound pro strategy (cf. Cinque 1990) may not
play a role in shaping the peculiar behavior of Chinese wh's-in-situ, and
unselective binding should be identified neither with (resumptive) pronominal
binding nor with generalized control in Huang's (1984) sense. Evidence from
specificity, inanimacy, and double island effects has testified the substantial
discrepancy between unselective binding and A'-pro binding.
As we proceed to consider the asymmetries among Chinese wh-adjuncts,
as well as those between syntactic and LF operations, it becomes clear that our
proposals, though simplistic at first glance, do cover a wide range of data
without tolling on the original insight behind the LCH. We demonstrate that the
seemingly random distribution and interpretations of Chinese wh-adjuncts are
actually systematic manifestation of their nominality and certain fundamental
properties of binding.
In the next chapter, we will take a step further into the border between
syntax and semantics.

122
CHAPTER THREE

TOWARD LF INTERFACE

0. 3-Closure Extensions

One of the remarkable discoveries concerning Syntax-Semantics interface


is that syntactic structures can be closely related to their logical representations
by a fairly explicit mechanism, as stated below in Diesing (1992a,b):

(1) Mapping Hypothesis:


a. Material from VP is mapped into the nuclear scope.
b. Material from IP (excluding VP) is mapped into a restrictive clause.

Diesing demonstrates that, by splitting a syntactic tree in correspondence with


the tripartite representations developed by Kamp (1981) and Heim (1982)
(consisting of a quantifier, a restrictive clause, and a nuclear scope), the Syntax-
Semantics mapping can be implemented in a straightforward manner, as
illustrated the following derivation: 1

(2) [IP Every donkey [vP kicks a man ]


Vx [x is a donkey] 3y y is a man Ax kicks y
quantifier restrictive clause 3-closure nuclear scope

1 Based on Lewis's (1975), Kamp (1981) and Heim (1982) propose that
indefinite NPs are not intrinsic quantifiers. Rather, they introduce variables,
which, in absence of other potential binders, are licensed by a default
existential operator. Diesing (1992), on the other hand, contends that there are
two types of indefinites: quantificational vs. non-quantificational. Quantificational
indefinites undergo Quantifier Raising (QR), resulting in presuppositions, while
non-quantificational ones stay in situ, licensed by 3-Closure.
123
She further argues that this approach sheds light on a number of mysteries
around the issue of specificity, whose accounts crucially relies on the
assumption that the Existential closure (3-closure) closes off VP, rather than IP.2
A significant consequence of the proposal is that any variable outside the
nuclear scope has to be licensed in a marked way, e.g., by strong quantifiers, by
generic tense, or by sentential adverbials such as always and usually. It follows
that the IP Spec and the VP Spec have different status in regard to the
interpretation of indefinites: An indefinite in the IP Spec must be quantificational,
whose interpretation is determined by the type of operator available in a given
sentence. On the other hand, an indefinite in the VP Spec is licensed by 3-
closure, yielding a non-specific exitential reading. In this case, the indefinite is
treated as a cardinal predicate within the nuclear scope (cf. (2)). The IP-VP
distinction is backed by two S-structure positions for German subjects, each of
which corresponds to one of the readings mentioned above. Take the bare
plural Linguisten 'linguists' in (3) for example (data from Diesing 1992a):

(3) a. .. . weil ja doch [vP Linguisten Kammermusik spielen].


since Prt Prt linguists chamber music play
... since there are linguists playing chamber music.

b. .. . weil [Ip Linguisteni ja doch [vp ei Kammermusik spielen]l].


since linguists Prt Prt chamber music play
... since (in general) linguists play chamber music.

2 Heim's (1982) original proposal is to associate 3-Closure with the nuclear


scope, roughly corresponding to the syntactic category IP. 3-Closure, in this
sense, is strictly a matter of semantics. As for the claim to confine 3-Closure to
VP, a similar proposal has been made by Higginbotham (1985:561) on a quite
different ground to define the domain of existential generalization over event
arguments (cf. Davidson 1967).
124
As Diesing points out, the bare plural in question is associated with a cardinal
(non-specific existential) reading when located lower than the particles ja and
doch, presumably in the VP Spec of (3a). In contrast, the bare plural is
interpreted as generic (i.e., quantified over by in implicit generic operator)
when located higher than ja and doch, presumably in the IP Spec of (3b) aftei
raising. The contrast thus follows directly from the mapping hypothesis (1) plus
the assumption that 3-closure applies only to the VP node.
Given the elegant way the theory works out, one may wonder whether the
whole conception of the Syntax-Semantics mapping can be further extended to
the topic-comment configuration, in view of the cross-linguistic generalization
that a topic has to be definite or generic. This point is illustrated by the contrast
between (4a,b) in English and that between (5a,b) in Chinese: 3

(4) a. John, I like.


b.* A man, I like.

(5) a. Akiu, wo hen xihuan.


Akiu I very like
'Akiu, I like very much.'

b.* yi-ge ren, wo hen xihuan.


one-CL person I very like
'*A man, I like very much.'

Furthermore, a bare plural is disambiguated when it is topicalized, as shown by


the contrast between (6) and (7):

3 A pointed out by Orin Percus (p.c.), (4b) is acceptable when a man is


construed as "a certain man", a specific interpretation unavailable for its
Chinese counterpart in (5b). We will see in the later discussion that it is this
asymmetry which allows us to single out the factors behind specificity effects,
and to seek out a more accurate characterization of specific indefinites.
125
(6) I always salute firemen.
a. Alwaysx [x is a fireman] I salute x
b. Always t [t is a time] 3x (x is a firemen AI salute x)

(7) Firemen, I always salute.


a. Alwaysx [x is a fireman] I salute x
b. #Always t [t is a time] 3x (x is a firemen AI salute x)

Both the quantificational reading (6a) and the cardinal reading (6b) are
available when the bare plural firemen remains in the object position at S-
structure. 4 In contrast, the quantificational reading survives the cardinal one in
(7), where the bare plural stands in the topic position.
As a starting point, we would like to adopt Chomsky's (1977) view that
topicalization involves a base-generated topic and a comment clause
containing an empty operator, as illustrated below:

(8) Topic, [comment Opi[ ... ei ... ]].

The raising of the empty operator creates an open sentence, which in turn is
predicated on the "topic-in-situ". The role of the empty operator can be
compared to that of a lambda operator in formal semantics. Now let's assume
that the empty operator undergoes IP-adjunction in the sense of Lasnik & Saito
(1992). An account immediately suggesting itself would be to assimilate the
topic-comment structure to the subject-predicate structure in the following terms:

(9) a. Material from IP is mapped into the nuclear scope.


b. Material from CP (excluding IP) is mapped into a restrictive clause.

4 In (6a), firemen is adjoined to IP at LF under QR (quantifier raising), and


accordingly mapped into the restrictive clause. For detailed discussions of
indefinite objects and problems caused by obligatory QR in generic context, see
Diesing (1992).
126
As a result, (7) is correctly mapped into the logical representation (7a), where 3-
closure would apply vacuously, and hence does not apply to avoid vacuous
quantification, as sketched by (10):

(10) [CP Firemen, [ip Opi [lp I always admire eil]


Alwaysx [x is a fireman] Xy (*3) 1admire y
quantifier restrictive clause 3-closure nuclear scope

At first glance, the mapping hypothesis appears to be trivialized under our


approach. In the following discussion, we will show that (1) and (9) can be
made to follow from a more general algorithm, according to which 3-closure is
associated with the notion of "nuclear scope", rather than a specific category
like VP, while the syntactic corespondent of a nuclear scope is locally defined
by the notion of a "syntactic predicate" in a cyclic manner, as sketched below:

(11) Extended Mapping Hypothesis (EMH):


a. Material from a syntactic predicate is mapped into the nuclear scope.
b. Material from the immediate dominating XP of a syntactic predicate
(excluding that predicate) is mapped into a restrictive clause.

A syntactic predicate is defined as a (one-place) predicate inducing predication


rather than modification.
Section 1 will spell out some general problems which have to be dealt with
by anyone who commits herself/himself to the mapping hypothesis. Particularly,
we will show that the mapping mechanism is a sort of cyclic operation,
presumably working side by side with semantic composition. In section 2, we
will show that there are substantial reasons to return to Heim's (1982) original
position, i.e., associating 3-closure with the nuclear scope. This move opens the
otherwise closed class of candidates which might be mapped into the nuclear

127
scope (according to Diesing's formulation, there is only one member, namely,
VP). On empirical grounds, we will show that non-restrictive relatives and a
certain type of secondary predicate display crucial characteristics of the
comment clause in (8): they count as syntactic predicates and induce specificity
effects on the subject of predication. We will then propose to reformulate the
mapping hypothesis in a broader term. Section 3 demonstrates that linguistic
variations of indefinite construals, as well as the distinction between individual-
level and stage-level predicates, can be accounted for in terms of the copying
mechanism developed by Chomsky (1992), without resorting to lowering and
VP-external subjects. The claim is that languages (and predicates) differ in the
possibility of leaving copies in chain formation: only when a copy is left in the
VP Spec can the relevant subject be interpreted within the nuclear scope, and
hence licensed by 3-closure.

128
1. Mapping as a Cyclic Operation
1.1. The Cyclic Hypothesis

A notable problem with the mapping hypothesis (1) is that it is not entirely
clear how it works when a bi-clausal structure is involved, as illustrated below:

(12) IP1
Spec I'
INFL VP1
Spec ...... IP2
Spec I'
INFL VP2

There are three logical possibilities. The first one is to implement the mapping in
a top-down manner, which we will reject on empirical grounds, since it wrongly
predicts that a variable in the IP2 Spec can be licensed by 3-closure on the VP1
node in (12). As evidenced by (13), embedding a clause with an individual-level
predicate like admirable does not add to its subject, i.e., firemen, an extra
cardinal reading: 5

5 Carlson (1977) distinguishes stage-level predicates like available from


individual predicates like admirable: Only the former allows a bare plural
subject ambiguous between the generic reading, as in (ia), and the cardinal
reading , as in (ib,c):
(i) Firemen are available.
a. Genx,t [x is a fireman At is a time] x is available at t
b. Gent [t is a time] 3x x is a fireman Ax is available at t
c. 3 x x is a fireman Ax is available
In contrast, a bare plural subject can only be generic in the presence of an
individual-level predicate, as shown by (ii):
(ii) Firemen are admirable.
a. Genx [x is a fireman] x is admirable
b. #3x x is a fireman Ax is admirable
129
(13) [Ip I [vP think [ip firemen [VP are admirable]]]].

An even more robust argument comes from Chinese indefinite subjects. As


Cheng (1991) observes, the well-known specificity/definiteness restriction on
Chinese subjects (cf. Li & Thompson 1981, Lee 1986, among others), as shown
by the contrast between (14) and (15a,b), is best explained by adopting the
mapping hypothesis, given that Chinese indefinites, unlike their English
counterparts, lack intrinsic quantificational properties, and the strategy of
lowering is not available in Chinese:

(14) * yi-ge ren zou-le.


one-CL person leave-Prf
'A person left.'

(15) a. you yi-ge ren zou-le.


have one-CL person leave-Prf
'There is a person who left.'

b. na-ge ren zou-le.


that-CL person leave-Prf
'That person left.'

Specifically, (14) is ruled out because it contains an unbound variable


introduced by yi-ge ren 'a person', which is outside the VP and hence beyond
the reach of 3-closure. On the other hand, the variable in question is licensed by
the existential predicate you 'have' in (15a), and by the demonstrative na 'that'
in (15b). 6 Now consider the following sentence:

For detailed discussions of how individual-level predicates can be


distinguished from stage-level predicates in syntactic terms, see Kratzer (1989)
and Diesing (1992).
6 Following Huang (1987,1989a), Cheng (1991:131) treats you 'have' as an
auxiliary, and further spells out its status as a modal heading a functional
130
(16)* [Ip Akiu [vp shuo [Ip yi-ge ren [vP zou-le]]]].
Akiu say one-CL person leave-Prf
'Akiu said that a person left.'

Recall that, unlike English, there is no way to restore the embedded subject yi-
ge ren to the lower VP Spec. Neither is the so-called presuppositional reading
available, since Chinese indefinites are [- quantificational force] in Diesing's
sense. If the mapping algorithm works in a top-down manner, the embedded
subject should get licensing from 3-closure on the matrix VP. This prediction,
however, is falsified by the deviance of (16), which is parallel to that of (14).
Another possibility is to map VP1 and VP2 simultaneously. As noted by
Irene Heim (p.c.), this move inherits the same problem as the top-down
approach. Namely, we wrongly predict that the embedded subjects in (13) and
(16) can be licensed by 3-closure on the matrix VP.
The only option left is to implement the mapping cyclically, which, in my
opinion, is the correct one. Let's start with (14). Bottom-up-wise, the first IP-VP
pair is mapped to (17) according to (1):

(17) ....[iP firemen [vP are admirable]] --


Genx [x is a fireman] x is admirable

projection higher than AspP. This move allows yi-ge ren 'a person' to be
accommodated by the AspP Spec in (15a). Yi-ge ren is thus existentially
quantified due to the assertion by you, yielding the cardinal reading. On the
other hand, there is also some evidence suggesting that you may serve as a
(strong) determiner, as shown by (i):
(i) you-de ren zou-le.
have-PNM person leave-Prf
'Some of the people left.'
The interpretation of you-de ren in (i) is clearly presuppositional. And the
presence of -de, a prenominal modifier marker (PNM), indicates that you is a
specifier rather than a higher predicate.
131
The variable introduced by firemen is bound by a generic operator triggered by
the generic tense (cf. Wilkinson 1986), and the indefinite itself is mapped into
the restrictive clause. On the other hand, since admirable counts as an
individual-level predicate, whose subject cannot be restored to the VP Spec in
Syntax, the following mapping is blocked: 7

(18) .... [Ip firemen [vP 3 [vp are admirable]]] -x--


3x x is a fireman Ax is admirable

Since the mapping is done in a local manner, 3-closure due on the next cycle
cannot affect the scope interpretation of firemen. This move thus correctly
predicts the lack of the cardinal reading in (14). Similar situations obtain for
(16), as illustrated below:

(19) .... [IP yi-ge ren [vp 3 [vP zou-le]]] -x--


one person left
3x x is a person Ax left

(20) .... [IP yi-ge ren [vP 3 [Vp zou-le]]] --


one person left
* [x is a person] 3x x left

As shown by (19), the subject yi-ge ren cannot be mapped into the nuclear
scope, and therefore is not eligible for licensing from 3-closure. When yi-ge ren
does get mapped to the restrictive clause, as in (20), a variable is introduced,
but no quantifier is available. Since the operation observes cyclicity (and

7 Here we do not necessarily commit ourselves to any particular theory of


the stage/individual asymmetry, such as Diesing's (1992a,b) proposal that an
individual-level predicate takes PRO as its VP-internal subject, while an
external lexical subject is independently 0-marked by its INFL. In section 3, I will
explore an alternative in terms of Chomsky's (1992) copy theory. At this stage, it
suffices to asst:. e that the subjects of individual-level predicates, just like
Chinese subjects, never take scope positions under the local VP node.
132
applies derivationally; see below), nothing from the next cycle may rescue the
violation. (16) is then ruled out in the same way as (15a).
A more interesting case comes from toric constructions containing an
indefinite subject, as exemplified by (21a,b):

(21) a.* [cP Akiu, [IP Op. [ip yi-ge nuOhai [vP hen xihuan ei]]]].
Akiu one-CL girl very like
'?Akiu, a girl likes.'

b. [cp Akiu, [Ip Opi [IP you yi-ge nUhai [vP hen xihuan eill]l.
Akiu have one-CL girl very like
'Akiu, there is a girl who likes (him).'

As a working hypothesis, we add (9) into the inventory of mapping principles,


and tentatively assume that 3-closure is associated with the nuclear scope
rather than VP. Let's start with the lowest cycle of (21a) according to (1):

(22) .... [Ip Opi [IP yi-ge nOhai [vP hen xihuan ei 1]1]
one girl very like
*%x [yisagirl] 3y y likes x

Again, an unbound variable is created in the restrictive clause. Now even if the
mapping proceeds to the next cycle according to (9), 3-closure on the relevant
nuclear scope (corresponding to IP) cannot license the offending variable, as
indicated by the deviance of (21a). In other words, the following mappings
should in principle be ruled out:

(23) [cP Akiu, [Ilp Opi [!P yi-ge nOhai [vP hen xihuan ei]]]]
Akiu one girl very like
Akiu, hx (3y [yis a girl] ylikesx) or
Akiu, x ( 3y [y is a girl] 3y (y likes x))

133
To achieve this, we have to implement the mapping cyclically, and decide its
legitimacy in "real time", i.e., derivationally rather than representationally. On the
other hand, since (21b) contains the existential predicate you 'have', which
contributes quantificational force to the indefinite subject, the variable in
question is licensed, and no 3-closure applies in either cycle:

(24) [cP Akiu, [ip Opi [IP you yi-ge nOhai [VP hen xihuan eil]]] -
Akiu have one girl very like
Akiu, x (3y [y is a girl] y likes x)

As a matter of fact, this derivational view of mapping also entails a bottom-


up-style composition. As we take a piece of syntac,,. structure into the
machinery, the resulting representation automatically becomes the building
block for another cycle of composition. When mapping succeeds, composition is
also accomplished. The problem is how big a chunk we should take for a cycle
of mapping-composition. The answer, in our opinion, lies in the notion of a
syntactic predicate (incontrast to a modifier/restrictor), which, as an unsaturated
function in Frege's sense, is ready to play the argument-taking role in
composition. We may therefore define a mapping cycle as a domain containing
a syntactic predicate and its subject (or whatever it predicates upon), of which
the IP in (1) and the CP in (9) are only two outstanding examples. The nuclear
scope, accordingly, is the domain of a syntactic predicate, as the VP in (1) and
the IP in (9) represents. In section 2.1, we will show how the proposal fits into
our characterization of 3-closure, which in turn leads to our formulation of (11).

134
1.2. The Mapping Geometry of Topics

A fair comment on the theory presented so far concerns that, although it


seems justified to say that the mapping mechanism (1) should apply cyclically,
extending the same analysis to topic-comment constructions may push things
too far. As Hubert Truckenbrodt and Orin Percus (p.c.) points out, a topic and a
raised subject have the same status in regard to the mapping geometry.
Namely, since both of them are located higher than VP, it is expected that they
should be mapped into their respective restrictive clauses, manifesting similar
specificity/definiteness effects. If topicslization is an instance of IP-adjunction
(Baltin 1982, Lasnik & Saito 1992), (1) correctly predicts the parallel between
subject-predicate and topic-comment constructions without resorting to the
seemingly redundant statement (9).
However, there is a catch in this picture: it focuses on the similarity shared
by topics and subjects, while a crucial distinction between them is missed. That
is, a (discourse) topic can never be interpreted as cardinal, but a subject may or
may not, depending on its ability to be restored to the VP Spec. From this point
of view, the base-generation/non-movement hypothesis (8) is preferred to the
adjunction hypothesis. As shown by (25), the base-generation hypothesis in
principle rules out the cardinal reading (i.e. the possibility of being quantified by
3-closure), since there is no way to restore the "topic-in-situ" back into VP:

(25) CP
Topic C'
C IP
Opi IP
Subj ... ti...

135
On the other hand, some stipulation has to be made for the adjunction
hypothesis to prevent a topic from lowering (Diesing 1992a) or leaving a copy
(Chomsky 1992). It is instructive to note that our observation does not
necessarily go against the adjunction/movement analysis. In the first place, all
the so-called topic island effects observed by Lasnik & Saito (1992) can be
attributed to the operator adjunction in (25), as exemplified by (26a-d):

(26) a.* Wherei did, that book, [ip Opj [1P you put tj ti
]]?
b.* She wonders where ,that book, [IP O pj [lp you put tj till]].
c.* I cleaned the place wherei, that book, [Ip Opj [lp you put tj ti ]].
d.* On the tablei, that book, lip Opj [ip you put Ljti
]].

Besides, there is certainly no point to reconstruct an empty operator, just as we


expect of the interpretation of topics. In the second place, it is widely noted that
there are two types of topic (cf. Gundel 1974, Kitagawa 1982, Culicover 1991).
One is the "discourse" topic, typically associated with the specificity/definiteness
effects under discussion here. The other is often called the "contrastive" or
"focus" topic. It has been proposed by Culicover (1991) that there are two
distinct landing sites for them: the IPadjunction site for a discourse topic, and
the Spec of Pol(arity)P for a focus topic. The topic island effects mentioned
above are weakened when we stress the topics and eliminate the pauses:

(27) a. ??She wonders wherei THAT BOOKj you put tjti.


b. ?? I cleaned the place wherei THAT BOOKj you put tjti.
c. ??On the tablei THAT BOOKj you put tjti.

Since it is substitution (i.e., raising to the PolP Spec) rather than adjunction that
applies to the topic in question, A'-movement is not blocked in (27a-c). Another

136
important distinction comes from the fact that a focus topic induces weak
crossover effects (cf. Lasnik & Stowell 1991, among others), while a discourse
topic does not, as shown by the following contrast (data from Culicover 1991):

(28) a. Robini, [iP Opi [Ip hisi mother really appreciates ti]].
b.* ROBINI hisi mother really appreciates ti.

The absence of weak crossover effect in (28a) follows from our account
straightforwardly: Since an empty operator does not bear referential
dependency, whose major role is to mediate the identification of its trace
(Chomsky 1986), no weak crossover configuration is formed. On the other hand,
a focus does involve quantification of some sort (Chomsky 1977, Rooth 1985),
and hence the weak crossover violation of (28b).
It is also possible to tease them apart in terms of indefinite interpretations.
For example, one can say something like (29) in a perfectly good sense,
provided that a contrast is made in the discourse:

(29) ONE PAPER, I can handle. (But two papers, that's too much.)

The topic in (29) is construed as cardinal, most naturally under stress. This
suggests that the contrastive topic is really a focus, which undergoes raising to
the Spec of F(ocus)P (or PolP inCulicover's sense), as illustrated below:

(30) FP
Objk F'
F IP
Subji I'
I VP
tk ...ti...

137
The discourse/focus distinction is even more clearly defined in Chinese.
When a topic appears higher than a subject, both the discourse and contrastive
construals are allowed, as shown by (31a) and (31b):

(31) a. zhe-pianryi-pian lunwen, wo hen xihuan.


this-CL/one-CL paper I very like
'This paperra paper, I like very much.'

b. yi-pian lunwen, wo hai keyi yingfu. (liang-pian, na jiu tai duo le.)
one-CL paper I still can handle two-CL that then too much Prt
'One paper, I still can handle. (Two papers, that's too much.)'

On the other hand, when a topic appears lower than a subject, the contrastive
construal is obligatory:

(32) a. wo zhe-pian lunwen xihuan, *(na-pian lunwen bu xihuan).


I this-CL paper like that-CL paper not like
'This paper, I like, but that paper, I don't.'

b. wo yi-pian lunwen keyi yingfu, *(liang-pian jiu bu xing le).


I one-CL paper can handle two-CL then not capable Prt
'One paper, I can handle, but two papers, I am not capable
(of handling).'

It is thus unlikely that the SOV order in (32a,b) is a consequence of double


topicalization (i.e. with the subject and the object both topicalized). Otherwise
we would expect the presence of an alternative discourse construal and the
specificity effect shown by (31a). Rather, it is the object which undergoes focus
movement, presumably into the Spec of an FP projection between IP and VP,
as illustrated in (33): 8

8 It is also suggested by Culicover (1991) that, in English, there is an


alternative PolP projection similar to FP in (33). One of its major functions is to
host wh- or negative subjects in matrix clauses, which is supposed to account
for the lack of subject-aux inversion in the same environment.
138
(33) IP
Subjk I'
I FP
Obji F'
F VP
tk t i...

Our position is supported by the fact that a similar word order is also observed
infocus constructions like (lian)... ye'(even) .. . also':

(34) a. Akiu (lian) zhe-dian qian *(ye) yiao.


Akiu even this-small-amount money also want
'Akiu even wants this (small) amount of money.'

b. Akiu (lian) yi-mao qian *(ye) yiao.


Akiu even one-penny money also want
'Akiu even wants one penny.'

As shown by (34a,b), only in the presence of focus markers like ye 'also' is the
object allowed to front, resulting in the SOV order.9

9 A more up-to-date version of our view can also be built upon Koizumi's
(1994) layered specifiers analysis, as illustrated below:
PolP
Topic PolP
Opk [+wh] Pol'
XPi [+F] Pol'
Pol [+F,+wh] IP
Subj ... tk. ti...
139
In sum, there seems to be a fundamental distinction between a discourse
topic and a focus topic, not only in their syntactic behavior, but also in their
mapping geometry. If our analysis is on the right track, then a discourse topic
should not be included in the IP-VP cycle, but a focus topic should (here we
regard FP as a "split" part of IP (Pollock 1989), whose head bears a V-feature in
Chomsky's (1992) sense). Consequently, we correctly predict that focus topics
do not display the specificity effects typically associated with discourse topics,
since they are eligible for licensing from 3-closure under reconstruction.

1.3. Summary

We have demonstrated that there are some merits to be gained if we


conceive of the syntax/semantics mapping as a cyclic operation. Not only is the
local nature of 3-closure derived in a principled way, but the parallel between
subject-predicate and topic-comment constructions also follows, provided that
we associate 3-closure with the nuclear scope instead of VP, and accordingly
allow (9) to play along with (1) during mapping. This move, in turn, leads us to
the conjecture that the mapping hypothesis has a wider range of application
than previously thought. In the next section, we will address this issue, and
present a general picture of where the mapping mechanism should apply.

140
2. 3-closure as a Post-Cyclic Operation

2.1. The Nature of 3-closure

In Diesing's (1992a) original conception, the domain of VP and the nuclear


scope appears to be synonymous with respect to the application of 3-closure. In
this paper, we would like to contend that there are some differences between
saying that 3-closure closes off VP and saying that 3-closure closes off the
nuclear scope. Only with the latter is the addition of (9) possible. By claiming so,
however, the role of 3-closure is somewhat blurred: it is implied that 3-closure
applies on logical representations assuming tripartite structures), rather than on
their syntactic counterparts (assuming X'-structures), as originally proposed by
Heim (1982). We are thus obliged to examine the nature of 3-closure more
closely.
The first notable property of 3-closure is that it does not interact with other
types of quantification. As demonstrated by Diesing, the following multiple-
quantified sentence actually has three distinct logical representations:

(35) Every cellist played some variations.


a. Everyx [x is a cellist] Somey [y is a variation] x played y
b. Somey [ y is a variation] Everyx [x is a cellist] x played y
c. Everyx [x is a cellist] 3y y is a variation Ax played y

In the first two interpretations, the object is quantificational. Namely, it


undergoes QR (quantifier raising), and interacts with the subject, resulting in the
scope ambiguity: every cellist takes scope over some variations in (35a), while
some variations takes scope over every cellist in (35b). Both interpretations
presuppose the existence of a set of variations: (35a) should be read as 'Every
cellist played some of the variations'; (35b) should be read as 'There are some

141
variations which every cellist played'. These are so-called presuppositional
readings for some. In contrast, the object is not quantificational in (35c), whose
quantificational force is contributed by 3-closure. Accordingly, some is read as
cardinal, and there is no pre-established set of variations. The fact that (35) can
not be construed as (36), which is roughly a logical equivalent of (35b) without
the presupposition, indicates that the existential quantifier introduced by 3-
closure can never be QRed:

(36) 3y Everyx [x is a cellist] y is a variation Ax played y or


3F Everyx [x is a cellist] F(variation) Ax played F

3-closure thus has a "post-cyclic" quality, applying only after other types of
quantification have applied, as the term "closure" has already implied. Although
associating 3-closure with VP does characterize its narrow-scope-taking
behavior, the stipulation itself does not dictate the peculiar property mentioned
above. On the other hand, if we associate 3-closure with the nuclear scope, the
post-cyclic quality follows straightforwardly, since 3-closure is introduced only
after the mapping applies, and therefore not subject to QR during the mapping
from S-structure to LF.
Second, 3-closure does not induce relativized minimality effects. This point
can be illustrated by the dual readings of Chinese wh's-in-situ, i.e., indefinite vs.
interrogative. As noted by Cheng (1991) and Li (1992), among others, Chinese
wh's-in-situ, in parallel to Chinese indefinites, do not possess inherent
quantificational force. First compare (37a) and (37b,c,d):

(37) a. Akiu chi-le shenme (ne)?


Akiu eat-Prf what Q
'What did Akiu eat?'

142
b. Akiu mei chi shenme.
Akiu have-not eat what
'Akiu didn't eat anything.'

c. Akiu dagai/keneng chi-le shenme.


Akiu probably/possibly eat-Prf what
'Akiu probably/possibly ate something.'

d. Akiu shenme-dou chi.


Akiu what-all eat
'Akiu eats everything.

While shenme 'what' gets the expected interrogative reading in (37a), it


behaves like a negative polarity item in (37b), and is existentially quantified in
(37c) and universally quantified in (37d). To take a c!oser look, we will start with
Cheng's proposal, where shenme 'what' is treated as a polarity item in all four
cases. In (37a), the optional wh-question marker ne and its empty counterpart
are both triggers and binders, so is dou 'all' in (37d). On the other hand,
negation and modality operators are triggers in (37b,c) respectively, while 3-
closure provides binders in both cases. 10 We may thus present the logical
representations of (38a-d) as follows:

(38) a. ?x [x is a thing] Akiu ate x


b. - 3 xis a thingA Akiu ate x
c. 3x x is a thing A Akiu ate x
d. Vx [x is a thing] Akiu eats x

o10 Cheng (1991) did not exactly talk about cases like (37c). However, there
are some suggestive clues in her treatment of free choice any (cf. Ladusaw
1979). According to Cheng, free choice any is treated exactly like its polarity
counterpart, with modality operators like will in (i) serving as both a trigger and a
binder (contributing the universal reading of any):
(i) Any Chinese semanticist will tell you that a white horse is not a horse.
Dagai 'probably' and keneng 'possibly' in (37c) may well behave similarly to the
English modal, except that they perform a single role as triggers.
143
Now let's consider a rarely explored fact: (38b) is not the only reading of
(37b), which can also be construed as a wh-question, as evidenced by the
presence of the Q-marker ne in (39). The same observation applies to (37c), as
shown by (40):

(39) Akiu mei chi shenme (ne)?


Akiu have-not eat what Q
'What did Akiu not eat?'

(40) Akiu dagai/keneng chi-le shenme (ne)?


Akiu probably/possibly eat-Prf what Q
'What did Akiu probably/possibly eat?'

In contrast, shenme 'what' cannot be construed as existentially quantified in


(37a), or as interrogative in (37d), as illustrated by (41) and (42) respectively: 11

(41) * Akiu chi-le shenme.


Akiu eat-Prf what
'Akiu ate something.'

(42) * Akiu shenme-dou chi (ne)?


Akiu what-all eat Q
'What did Akiu eat all (the time)?'

11 The exact syntactic position of dou is still under debate, due to its clitic
nature and various interpretations (at least including 'all', 'also', 'already', and
'always'). In the case of quantification over wh's-in-situ, dou takes scope over
the immediate constituent to its left, serving as an universal unselective binder.
As a matter of fact, if a wh-in-situ is not in the scope of dou, the interrogative
construal is possible. This point can be made clear by comparing (i) with (42):
(i) Akiu (pingchang) dou chi shenme (ne)?
Akiu usually all eat what Q
'What does Akiu eat all (the time)?'
As shown by (i), when shenme is not fronted to the left of dou, the question
reading emerges, with dou quantifying over time rather than individuals. For
further discussions, see Huang (1982), Lee (1986), Chiu (1990), Cheng (1991),
Tsai(1993).
144
Since wh's-in-situ are treated as polarity items under Cheng's approach, they
differ from indefinites in that 3-closure is available for the former only when
triggered by negation and modality operators, but no such trigger is required for
the latter. This accounts for the deviance of (41), since there is neither a trigger
for 3-closure nor a trigger-binder such as ne or dou.
A question comes to mind at this stage: Why doesn't 3-closure block the
interrogative construals in (39) and (40), just as dou 'all' does in (42)? There is
no a priori reason to discriminate between universal and existential quantifiers
in regard to relativized minimality, and saying that 3-closure is optionally
triggered only trivializes the original insight. A more promising answer, it seems,
lies in the timing of 3-closure. That is, if the relevant existential quantifier is
introduced along Syntax-Semantics mapping, then it is natural that it does not
induce minimality effects either in Rizzi's (1990) sense or in Aoun & Li's (1989)
sense. If there is no unbound variable in a logical representation after mapping,
as in the case of (39) and (40), 3-closure simply does not apply.
Alteratively, we may also take it as a working hypothesis that negative
operators such as mei 'have-not' and probability operators such as dagai
'probably' can be either strong or weak. 12 When they count as strong operators,
interrogative construals from higher scope positions are in principle blocked,
i.e., inducing so-called inner island effects. Furthermore, since mei and dagal
cannot by themselves serve as binders of indefinites, 3-closure must apply to
license the wh's-in-situ in question. 3-closure thus comes out as a last resort. In
case that meiand dagai do count as weak, the interrogative construal from the

12 Our discussion here is inspired by comments from Lisa Cheng, Jim Huang,
Toru Ishii, and Yuji Takano.
145
Q morpheme becomes possible, as in (39) and (40), and 3-closure remains as
a last resort when there is no other potential binder.
The latter approach raises an interesting conceptual problem within the
minimalist framework (Chomsky 1992): The crucial function of 3-closure is to
license unbound variables within the nuclear scope. Namely, it is not self-
serving, which goes against the grain of the Greed principle. 13 The conclusion,
however, is not forced if we reinterpret 3-closure's role in mapping theoretical
terms. Recall that we generalize the IP-VP split as formulated in (1) to a subject-
predicate split in section 1.1. The intuition which we pursue here is that 3-
closure is a disambiguation device guaranteeing that there is only one open
place per predication. It has been well established that the external O-role, if
any, is unique to a syntactic predicate (i.e., a predicate which triggers
predication rather than modification or restriction). A related observation is also
made by Napoli (1989), according to whom a predicate can have only one
subject role player. It follows that predication (in
contrast to 0-role assignment)
should in principle involve one-place predicates, in particular, one-place
complex predicates. 3-closure certainly works well to serve this purpose:

(43) a. [ip Subj(x)i [v ti ... Obj (y) ... ] ---*

QPx [Subj(x)] Xx Ey ( x... Obj(y) ...)

b. [cP Topic, [ip Op[Ip ... ti... ]]] --


QPx [Topic(x)] ;x (3)( ... x...)

As illustrated by (43a), there are two potential open places within VP: one is the
trace left by subject raising; the other is the variable to be introduced by the

13 This issue, as Noam Chomsky (p.c.) observes, would not arise if we treat
existential closure as some sort of interpretation procedure in the conceptual-
intentional system. Under this view, existential closure has neither syntactic nor
semantic status, and therefore is not subject to Economy considerations.
146
indefinite object. As mapping proceeds, the object variable is "sealed off" by 3-
closure, while the subject trace, already indexed, triggers the predication on the
raised subject. As for the topic-comment construction (43b), there is a one-to-
one correspondence between the empty operator and the lambda operator.
Since variable-licensing is settled on the previous cycle (i.e., within the domain
of IP, cf. section 1.1), any variable beyond IP has to be licensed in a marked
way. Hence the specificity /definiteness effects on topics.
Along this line, we may urify (1) and (9) under the assumption that 3-
closure closes off the nuclear scope, as we have proposed by (11):

(11) Extended Mapping Hypothesis (EMH):


a. Material from a syntactic predicate is mapped into the nuclear scope.
b. Material from the XP immediately dominating a syntactic predicate
(excluding that predicate) is mapped into a restrictive clatuse.

3-closure, under this particular formulation of the Mapping theory, is a self-


serving operation of a syntactic predicate, which ensures the success of
predication.
So far we have limited our discussion to the conceptual plausibility of the
EMH. The next step, therefore, is to examine its empirical consequences, to see
if the same analysis can be extended to other types of syntactic predicates. This
will be our main task for the following sections.

147
2.2. Non-restrictive Relative Clauses

A ready candidate for our inspection comes from non-restrictive


relativization, which shares a cluster of properties with topicalization. First, non-
restrictive relativization does not display weak crossover effects, as evidenced
by (44) and its Chinese counterpart (45):14

very much]], .....


(44) [DP The students], [cP whoi [iP theiri teachers like ti

(45) [DP naxie [cP Opi [IP tameni (-de) laoshi hen xihuan ti]] de xuesheng]
those their teacher very like PNM student

Second, the paired reading shown by the wh-question (46a) is not available for
relativization in (46b) and topicalization in (46c):

(46) a. Who likes whom?


b.* The couple [who likes whom]
c.* The couple, he likes her.

The contrast suggests that relative wh's and empty operators are not subject to
scope absorption in the sense of Higginbotham & May (1981). In other words,
they can not be "paired" in a particular CP domain. Third, neither non-restrictive
relativization nor topicalization display scope interaction effects. As noted by
May (1985), the following wh-question is ambiguous:

(47) What did everyone buy for Max?


a. What is x, x a thing, such that for every y, y a person, y bought x for Max?
b. For every y, y a person, what is x, x a thing, such that y bought x for Max?

14 I owe discussion here to Danny Fox, Martha McGinnis, and Colin Phillips.
See also Lasnik & Stowell (1991) for some
148
(47a) is the so-called collective reading, with what taking scope over everyone.
(47b) is the so-called distributive reading, with everyone taking wide scope. The
same type of scope interaction, however, cannot be found in non-restrictive
relative clauses and topic-comment structures, as shown by (48) and (49)
respectively:

(48) [DP The present], [whichi [everyone bought ti for Max]], .....
(49) [Top The present], [Opi [everyone bought ti for Max]].

In both (48) and (49), everyone can only be construed as collective, and no
ambiguity is detected. 15 We thus conclude that the operators involved in non-
restrictive relativization and topicalization are not quantifiers. Rather, we may
call them "predicate markers", whose function is to define the domain of a
complex predicate. This functional account, nevertheless, should be understood
as a rationale behind relativization and topicalization. The mechanical
execution of the operator movement, we believe, is achieved by morphological
checking in Chomsky's (1992) sense. In other words, the empty operator in (48)
and (49) should be treated like a relative wh, i.e., [+wh] but [-Q]. By checking its
features on a functional projection, say, the CP Spec, the operator is in a c-
commanding position to define the domain of a predicate, i.e., the relative
clause in (48) or the comment clause in (49).
Here an important question emerges: What is the semantic distinction
between (48) and (49) if both the non-restrictive relative clause and the

15 As pointed by Uli Sauerland (p.c.), the following example appears to allow


a distributive reading:
(i) A Tale of Two Cities, which everyone bought for Max on his birthday...
But note that here A Tale of Two Cities really refers to a collection of copies
rather than a single copy. In other words, it is construed as a kind in Carlson's
sense. A simiiar construal in restrictive relativization will be like (ii):
(ii)The kind of books which everyone bought for Max on his birthday...
149
comment clause are to be treated as predicates? Intuitively speaking, the
semantic type of (48) should remain the same as that of its head after
composition, i.e., an entity, whereas the composition of (49) should produce a
truth value. Therefore, they cannot both involve predication. As a matter of fact,
the relative clause in (48) behaves rather like an adjective associated with
modification, which in principle does not change the semantic type of its
"subject" (cf. Kamp 1975). The distinction, therefore, falls under the dichotomy
between modification and predication, as the classic discussion of the
difference between (50a,b) has already revealed:

(50) a. That is a big butterfly.


b. That butterfly is big.

As noted by Higginbotham (1985), the grading of bigness in (50a) is relative to


the average size of a butterfly, while the grading is rather open-ended in (50b).
Since a big butterfly could be small in comparison with other creatures, (50a)
could hold true for an object for which (50b) counts as false. (50a) may be
paraphrased as (51a) with the modifier big being treated as a conjunct, and
(50b) as (51b):

(51) a. That is a butterfly, and it is big (for a butterfly).


b. That butterfly is big (for an x).

As a comparison, Higginbotham further points out that there is another type


of adjectival usage which does not induce restriction, as in "Look at the little
butterfly". Since no assertion is made about the littleness relative to the average
size of a butterfly, little can be treated as an ordinary conjunct, as paraphrased
below:

150
(52) Look at the butterfly, and it is little.

What is involved here seems to be descriptive (based upon our world


knowledge that butterflies are little things) rather than restrictive, a quality closer
to (51b)than to (51a).
In parallel to the non-restrictive usage of adjectives, we also find a class of
relatives which induce no restriction on head nouns. They are traditionally
called non-restrictive or appositive relatives, often characterized by the
tendency to form an isolated intonation group, as exemplified by (54):

(53) Tt man who has a big head got stuck in the rnnholc
(54) The man, who has a big head, got stuck inthe manhole.

Although restrictive and non-restrictive relatives almost have the same syntactic
configuration as far as their internal structures are concerned, they do have
different status in regard to the mapping geometry. That is,the restrictive relative
in (53) should be mapped into a restrictive clause, while its non-restrictive
counterpart in (54) should be mapped into the nuclear scope, as illustrated by
(55a,b) respectively:

(55) a. Thex [x is a man Ax has a big head] x got stuck inthe manhole

b. Thex [x is a man] x has a big head Ax got stuck inthe manhole

it thus seems plausible to generalize the modification-predication dichotomy


further to the distinction between restrictive and non-restrictive relatives. As a
matter of fact, the same intuition has been pursued intensively in the literature.
For example, Jackendoff (1977) distinguishes an appositive from a restrictive
modifier by claiming that the former is a sisters of N", whereas the latter is a
sister of N'. Following Emonds (1979), Napoli (1989) treats appositives as

151
adjuncts under the I' node, which directly denies non-restrictive relatives their
modifierhood. Safir (1986), on the other hand, proposes that non-restrictive
relatives will not be attached to N"until a level after LF, i.e., LF', besed on a set
of binding puzzles typically associated with them (also cf. McCawley 1982). For
the purpose of this paper, we may tentatively treat a non-restrictive relative as a
conjunct of the main predicate. As a result, (53b) has the following structure:

(56) The mani, [cP whoi [1P ti has a big headj] and [vp tj got stuck in the manhole]

An outstanding prediction of the EMH in this respect is that an indefinite


subject cannot be construed as cardinal in the configuration of (56), since there
is no way to restore it into only one of the conjuncts (namely, the main
predicate) due to the across-the-board constraint. The other conjunct, i.e., the
non-restrictive relative, behaves exactly like a comment clause in topicalization,
allowing no reconstruction (cf. section 1.2). This prediction is borne out, as
evidenced by the contrast between (57a,b) with SOME standing for strongly
quantified (presuppositional) some, and Sm its weakly quantified (cardinal)
counterpart (cf. Milsark 1974, loup 1975, Barwise & Cooper 1981, Reuland
1983, De Jong & Verkuyl 1985, Higginbotham 1987, Reinhart 1987, Partee
1988, and Diesing 1992a, among others):

(57) a. SOME/Sm fool who has a big head got stuck inthe manhole.
b. SOMErSm fool, who has a big head, got stuck in the manhole.

The specificity effect associated with non-restrictive relatives is also found


in the case of indefinite objects, as shown by the following contrast:

(58) a. The gas company will fire SOME/Sm fool who has a big head.
b. The gas company will fire SOME*Sm fool, who has a big head.

152
The non-restrictive relative in (58b) behaves rather like a secondary predicate
(cf. Rothstein 1983), and can be paraphrased as a separate conjunct:

(58') The gas company will fire some fool, and (s)he have a big head.

As the paraphrase indicates, the subject of non-restrictive relativization has to


be specific (or familiar in the sense of Heim 1982 and En; 1991). This is exactly
the property which presuppositional some contributes to the object fools in
(58b). The strong quantified object (i.e., SOME fool) is therefore preferred to the
weakly quantified one (i.e., Sm fool ). Here the EMH again makes the right
prediction: Since only the non-restrictive relative in (58b) counts as a syntactic
predicate, it takes some fool as the subject, and together they form a mapping
cycle, as sketched below:

(59) IP
DPi ..... VP
I
the company ti V'
firedk VP - mapping cycle

subject -- DP V'

some fool tk CP - nuclear scope

who nas a big head

The non-restr'tive relative is located in an inner adjunct position within the


lower VP shell, serving as a complex predicate (cf. Larson 1988, Huang 1989b,
Hale & Keyser 1991). The effect of 3-closure thus shows up on the subject of
predication in the relevant cycle.(i.e., the lower VP node). It should be further
pointed out that the object some fool will not be exactly in the lower VP Spec at
LF either if we follow Diesing in claiming that presuppositional indefinites
153
obligatorily undergoes QR, or if we adopt Chomsky's proposal that object NPs
move covertly to the Spec of AGRO under morphological checking. As a result,
we are bound to consider a Chain an LF object (Chomsky 1991), and treat it
accordingly in mapping theoretical terms.
The next step of mapping then applies where the main predicate discharge
its external 0-role, i.e., the Spec of IP. In other words, IP is the next mapping
cycle, and I', the sister node of the subject company, is mapped into the nuclear
scope, as illustrated below:

(59') IP - mapping cycle

subject *- DPi I' -V nuclear scope

the company lo+firedk ..... VP


tj V'

tk

The restrictive relative in (58a), on the other hand, serves as a modifier in


DP projections. Consequently, it is mapped into the restrictive clause in the
case of presuppositional some, and stays in the nuclear scope when some fool
is licensed by 3-closure, as illustrated by (60a,b) respectively:

(60) a. Thex [x is a gas company] somey [y is a fool , 3z (z is a big head


A yhas z)] x fired y
b. Thex [x is a gas company] 3y y is a fool A3z (z is a big head
Ay has z) Ax fired y

It is also instructive to note that, within the relative clauses, there is another
cycle of mapping triggered by the predicate has a big head, during which 3-
closure applies to license the cardinal reading of a big head.

154
On empirical grounds, this "dynamic" version of mapping hypothesis is
further supported by Percus's (1994) observation that, in addition to unergative
subject indefinites, certain non-subject experiencer arguments also receive
obligatory quantificational readings. He correlates this peculiarity to a unique
thematic role of object experiencer predicates, which may well counts as
external and triggers predication on the experiencer argument in question.
Since no raising is involved (at least in the immediate mapping geometry), the
obligatory presence of specificity is expected under the EMH.
Moreover, Jonas & Bobaljik (1993) observes that a (non-specific) indefinite
subject always precedes a (overtly) shifted direct object in Icelandic, which
should be impossible under Diesing's approach. Bobaljik (1994) points out
further that this paradox calls for a more dynamic splitting similar to the one
dictated by the EMH.
If the above analysis is on the right track, then we should be able to find the
same patterns in Chinese/Japanese-type languages, where NP projections are
head-final, and relative clauses are uniformly located in prenominal positions.
First consider the following Chinese examples:

(61) a. Akiu zai-xie [DP na-ben [cp Opi [dajia dou hui xihuan ei ]] de shu].
Akiu Prg-write that-CL people all will like PNM book
'Akiu is writing that book which everybody will like.'

b. Akiu zai-xie [DP [CP Opi [dajia dou hui xihuan ei 1] de na-ben shu].
Akiu Prg-write people all will like PNM that-CL book
'Akiu is writing that book, which everybody will like.'

As shown by (61a), when a relative clause appears lower than a determiner,


i.e., the definite specifier na-ben, the reading is restrictive. It could be the case
that there are several books which Akiu is planning to writ, and (61a) asserts
that he is writing the would-be-popular one. In comparison, when a relative
155
clause appears higher than a determiner, as in (61b), the reading is non-
restrictive. What is asserted here is that Akiu is writing a book which is salient in
the context, and the book will be popular. 16 Now we substitute yi-ben 'one-CL'
for na-ben in (61a,b). The restrictive/non-restrictive asymmetry emerges (also cf.
Tang 1975, Hou & Kitagawa 1987), as shown by the contrast between (62a,b):

(62) a. Akiu zai-xie [DP yi-ben [cP Opi [dajia dou hui xihuan ei ]] de shu].
Akiu Prg-write one-CL people all will like PNM book
'Akiu is writing a book which everybody will like.'

b.* Akiu zai-xie [DP [CP Opi [dajia dou hui xihuan ei ]] de yi-ben shu].
Akiu Prg-write people all will like PNM one-CL book
'Akiu is writing a book, which everybody will like.'

The result follows straightforwardly from the EMH if we treat the non-restrictive
relative in (62b) as a syntactic predicate, and yi-hen shu 'a book' as its subject
within the DP cycle.17 Since the indefinite subject can be licensed neither by 3-
closure on the next cycle (cf. section 1.1) nor by 3-closure on the current cycle

16 But note that this restrictive/non-restrictive distinction is somewhat


controversial. See Chao (1968) and Huang (1982) for a completely different
view. We will return to resolve the difference in section 2.4.
17 See Hou and Kitagawa (1987) for an interesting ECP account of the above
asymmetry. Based on Tang (1975), they observe that there is a subject-object
asymmetry between (ia,b):
(i) a. [[ei dai yanjing] de neixierhenduo xueshengi] dou yonggong.
wear glasses PNM those/many student all hard-working
'[ThoserMany students [who wear glasses]] are hard working.'
b. [[wo renshi el] de neixie/henduo xueshengi ] dou dai yanjing.
I know PNM those/many student all wear glasses
'[Those/Many students [who I know]] wear glasses.'
They contend that the deviance only occurs when a subject is relativized in the
"outer" relative, as in (ia), due to standard ECP effects. And hence the
grammaticality of (ib). However, we find that their observation applies only in the
presence of factual verbs like renshi'know', which presuppose the existence of
their complements. The problem posed by (62b), therefore, does not go away
even if we assume that Chinese INFL does not count as a lexical governor, as
Hou and Kitagawa suggest. See section 2.3.1 for related discussion. Also
thanks to Akira Watanabe for bringing the above issue to my attention.
156
through reconstruction (cf. section 1.2), the sentence is ruled out. In contrast, the
restrictive relative in (62a), as a modifier, does not trigger predication. And no
definiteness effect is detected, just as expected. The same observation holds for
Japanese relative clauses, as exemplified below (data due to Hiroyuki Ura):

(63) a. ano [[ei ringo-o katta] Opi] hito


that apple-Acc bought person
'The person who bought apples'

b. [[ei ringo-o katta] Opi] ano hito


apple-Acc bought that person
'The person, who bought apples'

(64) a. san.-ni-no [[ei ringo-o katta] Opi] hito


thre'e-CL-Gen apple-Acc bought person
'Thiee persons who bought apples'

b.* [[ei ringo-o katta] Opi] san-ni-no hito


apple-Acc bought three-CL-Gen person
'Three persons, who bought apples'

As Masa Koizumi and Hiro Ura (p.c.) point out, the restrictive/non-restrictive
distinction between (a)- and (b)- clauses of (63) and (64) is parallel to that of
Chinese relative clauses. The same pattern of contrasts once again emerges
when non-restrictive relativization (and hence predication) is involved, as
shown by the deviance of (64b).
As a first approximation, we may analyze the restrictive relative in (62a) as
a sister of N', as illustrated in (65a), and the non-restrictive relative in (62b) as
an adjunct of DP, as illustrated in (65b):

157
(65) a. restrictive relative:
DP
D N'
yi-ben CP N'
Op IP shu
b. non-restrictive relative:
mapping cycle - DP

nuclear scope- CP DP -- subject

Op IP D N'
I I
*yi-ben shu
one-CL book

In (65a), the CP serves as a modifier, and no predication is involved. Therefore,


there is no mapping cycle in the DP except a trivial one within the IP node. On
the other hand, the CP in (65b) does trigger predication, and an independent
mapping cycle is formed according to the EMH. The subject in question, i.e., yi-
ben shu 'a book', cannot be restored back into the nuclear scope (the CP node).
Nor can it get licensed from within the current cycle (the higher DP node). The
mapping thus fails, and (62b) is correctly ruled out.
One thing has to be made clear before we leave this section. It is not easy
to characterize Chinese non-restrictive relatives in such a way that we can get
the desired semantics as in English: Since VP projections are essentially head-
first in Chinese (vs. head-final in Japanese), there is no inner adjunct position to
the left of head verbs. Consequently, while Japanese non-restrictive
relativization can be treated as the mirror image of what happens in (60), its
Chinese counterpart cannot. Something more, therefore, has to be said about
either the mapping mechanism itself or the empirical status of Chinese non-
restrictive relatives. We will delay the discussion till section 2.4.

158
2.3. Existential Constructions and Predication Licensing

2.3.1. Existential Predicates and (In)definiteness Restrictions

In our discussion of Chinese non-restrictive relativization, we tacitly


avoided use-type verbs like du 'read' (vs. "creation" verbs like xie 'write') since
they presuppose the existence of their objects (cf. Diesing 1992a). For example,
the contrast between (62a,b) is blurred once we replace zai-xie 'is writing' with
du-guo 'read (pst.)':

(66) a. Akiu du-guo [DP yi-ben [cP Opi [dajia dou hui xihuan ei ]] de shu].
Akiu read-Exp one-CL people all will like PNM book
'Akiu read a book which everybody will like.

b. ?Akiu du-guo [DP [CP Opi [dajia dou hui xihuan ei]] de yi-ben shu].
Akiu read-Exp people all will like PNM one-CL book
'Akiu read a book, which everybody will like.

To make a solid case out of the restrictive/non-restrictive asymmetry, we ought


to find a construction where strongly quantified NPs are not allowed. Existential
sentences with you 'have' are exactly what we need here:

(67) a. wo you yi-ge erzi.


I have one-CL son
'I have a son.'

b.* wo you na-ge/mei(-yi)-ge/da-duo-shu-de erzi.


I have that-CL/every(-one)-CLUmost son
' I have the/every/most son(s).'

You in (67a) expresses possession, and its subject is thematic. You may also
express existence. As Huang (1987) observes, existential you only allows non-
thematic subjects, which could be a locative nominal or simply an empty
expletive, as shown by (68a) and (69a) respectively:

159
(68) a. fangjian-li you yi-ge nQhai.
room-inside have one-CL girl
'In the room, there is a girl.'

b.* fangjian-li you na-ge/mei(-yi)-ge/da-duo-shu-de nihai.


room-inside have that-CL/every(-one)-CL/most girl
'* In the room, there is/are the/every/most girl(s).'

(69) a. e you yi-ge ntihai zai fangjian-li


have one-CL girl at room-inside
'There is a girl in the room.'

b.* e you na-ge/mei(-yi)-ge/da-duo-shu-de n ihai zai fangjian-li.


have that-CL/every(-one)-CL/most girl at room-inside
'* There is/are the/every/most girl(s) in the room.'

Furthermore, as indicated by the deviance of (b) clauses of (67-69), both


possessive you and existential you induce the (in)definiteness effects on their
(structural) objects. 18 For one thing, although the existence of the object erzi
'son' in (67) is not directly asserted by possessive you, it is entailed by the

18 A precaution here concerns that nominals allowing so-called "copy"


readings do not fall under the definiteness effects induced by you 'have', as
exemplified below:
(i) fangjian-li you na-ben/mei(-yi)-ben/da-duo-shu-de shu.
room-inside have that-CL/every(-one)-CL/most book
'In the room, there is (a copy of) the/every/most book(s).'
As noted by Huang (1987), DPs like na-ben shu 'that book' in (i) are
syntactically definite, but semantically indefinite. Huang further points out that
when the subject position is not filled by a lexical NP, even the copy reading is
ruled out.
(ii)* e you na-ben/mei(-yi)-ben/da-duo-shu-de shu zai fangjian-li.
have that-CL/every(-one)-CL/most book at room-inside
'There is (a copy of) the/every/most book in the room.'
The conclusion thus appears to be that, on the top of the semantic constraint
associated with you-sentences, as evidenced by (b)-clauses of (67-69), there is
a stronger syntactic constraint against an expletive subject linked to a definite
NP, which constitutes an unbalanced 0-chain in Safir's (1985,87) sense (also
cf. Safir 1982, Heim 1987, and Higginbotham 1987).
160
semantics of possession. Besides, (67a) does not seem to have the literal
meaning "I own a son". Rather, it should be read as "there is a son of mine".
Consequently, if we replace erzi with a stereotype of possessions like fangzi
'house', the contrast is weakened:

(70) a. wo you yi-dong fangzi.


I have one-CL house
'I have a house.'

b.? wo you na-dong/mei(-yi)-dong/da-duo-shu-de fangzi.


I have that-CL/every(-one)-CL/most house
'?I have the/every/most house(s).'

The deviance of (b) clauses is thus uniformly attributed to the incompatibility


between existential assertion and definiteness (cf. Barwise & Cooper 1981,
Huang 1987, among others), with the weakest violation when possession is
involved.19 If our analysis is on the right track, non-restrictive relatives should in
principle be ruled out in you-sentences. This prediction is borne out, as
evidenced by the following contrasts:

(71) a. wo you [DP yi-ge [cP Opi [ei xihuan nianshu]] de erzi].
I have one-CL like studying PNM son
'I have a son who likes studying.'

b.* wo you [DP [COp i[ei xihuan nianshu]] de yi-ge erzi].


I have like studying PNM one-CL son

19 At this stage, it is instructive to note that we largely follow Huang's (1987)


descriptive generalization except one point: While Huang distinguishes you-
sentences with empty expletive subjects like (69a) from the other two types of
you-sentences like (67a) and (69a) based on the contrast between (i) and (ii) of
footnote 10, we do not make the distinction and treat the contrast as a special
case (due to the "copy" reading), in view of the parallel between (67b) and (68b)
on the one hand, and (69b) on the other. As a matter of fact, this move can be
seen as a projection of Huang's general approach towards the definiteness
effects induced by Chinese existential predicates.
161
'1. I have a son, who likes studying.'

(72) a. fangjian-li you [DP yi-ge [cP Opi [ ei dai-zhe hong maodou]] de nOhai].
room-inside have one-CL wear-Dur red cloak PNM girl
'In the room, there is a girl who wears a red cloak.'

b.* fangjian-li you [DP [CP Op [ ei dai-zhe hong maodou]] de yi-ge n*hai].
room-inside have wear-Dur red cloak PNM one-CL girl
'. In the room, there is a girl, who wears a red cloak.'

(73) a. e you [DP yi-ge [cP Opi [ ei dai-zhe hong maodou]] de nUjhai] zai
have one-CL wear-Dur red cloak PNM girl at
fangjian-li.
room-inside
'There is a girl who wears a red cloak in the room.'

b.* e you [DP [CP Opi[ ei dai-zhe hong maodou]] de yi-ge n(*hai] zai
have wear-Dur red cloak PNM one-CL girl at
fangjian-li.
room-inside
'7. There is a girl, who wears a red cloak, in the room.'

As indicated by the deviance of (b) clauses of (71-73), non-restrictive relatives


are not allowed to predicate on the indefinites in you-sentences. Restrictive
relatives, on the other hand, are allowed to serve as their modifiers. The pattern
of contrasts of (71-73) is therefore correlated to that of (67-69), just as the EMH
predicts.
Another support of our conclusion comes from appearance/disappearance
verbs like lai'come', qu 'go', fasheng 'happen', and si 'die', which, as Huang
puts it, assert "coming into existence" or "yc(ng out of existence". Syntactically,
they also share the common property of being ergative. The (in)definiteness
effects thus come out strong:

162
(74) a. e lai-le liang-ge ren/*Lisi/*tana-ge ren/*me-ge ren le.
come-Prf two persons/Lisi/(s)helthat person/every person Prt
'(lit.) Came two personsfLisir(s)herthe personteverybody.'

b. e si-le liang-ge renrLisirtarna-ge renrme-ge ren le.


die-Prf two persons/Lisif(s)he/that person/every person Prt
'(lit.) Died two personsrLisi/(s)hefthe personteverybody.'

The restrictive/non-restrictive asymmetry, as expected, shows up promptly when


relativization is involved, as evidenced by the contrasts between (a) and (b)
clauses of (75,76):

(75) a. e lai-le [DP liang-ge [cp Opi [Akiu yaoqing ei ]] de ren] le.
come-Prf two-OL Akiu invite PNM person Prt
'(lit.) Came two persons who Akiu hates most.'

b.* e lai-le [DP [CP OPi [Akiu yaoqing ei]] de liang-ge ren] le.
come-Prf Akiu invite PNM two-CL person Prt
'(lit.) Came two persons, who Akiu hates most.'

(76) a. e si-le [DP liang-ge [cOpi [ ei bu xi-zao]] de ren] le.


die-Prf two-CL not take-bath PNM person Prt
'(lit.) Died two persons who don't take bathes.'

b.* e si-le [DP [COpi[ ei bu xi-zao]] de liang-ge ren] le.


die-Prf not take-bath PNM two-CL person Prt
'(lit.) Died two persons, who don't take bathes.'

So far we have shown that the definiteness effect induced by non-


restrictive relativization and that displayed by topicalization are of the same
nature. And they can be made to follow from the EMH given that predication is
involved in both cases, an assumption warranted on both conceptual and
empirical grounds.

163
2.3.2. Weak Existential Predicates and Secondary Predication

In addition to you'have' and (dis)appearance verbs, Huang (1987) further


point out two types of predicates which appear to be less existential in lexical-
semantic terms. One concerns "location verbs", including intransitives like zhu
'live', zuo 'sit', tang 'lie', you 'swim', and transitives like fang 'put' and gua
'hang', hua 'paint'. They are characterized by their selection of locative
arguments, and by their association with the durative aspect -zhe. (the
perfective aspect -le is also possible with the transitives). In contrast to the first
two types of existential predicates, the location verbs do not display the
(in)definiteness effect as a rule, as shown below:

(77) a. pingguo-shu-xia zuo-zhe yi-ge kexuejia/na-ge kexuejia/Niudun.


apple-tree-bottom sit-Dur one-CL scientist/that-CL scientist/Newton
'Under the apple tree sits a scientist/the scientist/Newton.

b. shui-li you-zhe yi-tiao meirenyO/na-tiao meirenyui/Zhubajie.


water-inside swim-Dur one-CL mermaid/that-CL mermaid/ZhuJajie
'In the water swims a mermaid/the mermaid/Zhubajie.'

c. qiang-shang hua-zhe/-le yi-wei pusa/na-wei pusa/Guanshiyin.


wall-top paint-Dur/Prf one-CL goddess/that-CL goddess/Guanshiyin
'On the wall is painted a goddess/the goddess/Guanshiyin.'

The same observation holds for action verbs associated with the experiential
aspect -guo and the perfective aspect -le :

(78) a. ta jiao-guo yi-ge xuesheng/na-ge xuesheng/Aiyinsitan.


(s)he teach-Exp one-CL student/that-CL student/Einstein
'(S)he has the experience of teaching a student/the student/Einstein.'

b. wo gu-le yi-ge zhentan/na-ge zhentan/Chen Chali.


I hire-Prf one-CL detective/that-CL detective/Charlie Chen

164
'I hired a detective/the detective/Charlie Chen.'

However, as noted by Huang, when we add a clausal predicate in the


sentence-final position, the (in)definiteness effect obtains without exception.
First compare (79a-c) with (77a-c) respectively. We find that once the secondary
predication is involved, location verbs behaves exactly like the typical
existential predicates. 20 In other words, the (in)definiteness effects re-emerge:

(79) a. pingguo-shu-xia zuo-zhe yi-ge kexuejia/?*na-ge kexuejia./?*Niudun


apple-tree-bottom sit-Dur one-CL scientist/that-CL scientist/Newton
[cP Opi[ e, hen xihuan chi pingguo]].
very like eat apple
'Under the apple tree sits a scientist/the scientist/Newton, who likes
eating apples.'
b. shui-li you-zhe yi-tiao meirenyPf?*na-tiao meireny0ciZhubajie
water-inside swim-Dur one-CL mermaid/that-CL mermaid/Zhubajie
[cpOpi[ ei hen piaoliang]].
very pretty
'In the water swims a mermaid/the mermaid/Zhubajie, who is very
pretty.'
c. qiang-shang hua-zhe/-le yi-wei pusafk.na-wei pusa/?Guanshiyin
wall-top paint-Dur/Prf one-CL goddess/that-CL goddess/Guanshiyin
[cP Opi [ ei jiaota lianhua]].
step-on water lily
'On the wall is painted a goddess/the goddess/Guanshiyin, who steps
on a water lily.'

20 Here we use the term "secondary predication" in a rather loose sense. In


fact, every instance of predication that we discuss so far counts as primary in
Rothstein's (1983) sense if we add CP to the inventory of primary predicates in
the spirit of the EMH, as stated oelow:
(i) Primary Predication:
A VP/CP bears a primary predication to an XP if
a. XP and VP/CP mutually m-command each other, and
b. XP binds an empty argument position in the VP/CP.
165
Action verbs inflected by -guo and -le also display the same pattern of
contrasts. This point can be illustrated by comparing (78a,b) with (80a,b)
respectively:

(80) a. ta jiao-guo yi-ge xuesheng/?Tna-ge xuesheng2,Aiyinsitan


(s)he teach-Exp one-CL student/that-CL student/Einstein
[cP Opi[ ei conglai bu jiao zuoye]].
ever not hand-in homework
'(S)he has the experience of teaching a student/the student/Einstein,
who never handed in homework.'

b. wo gu-le yi-ge zhentanP./na-ge zhentanf.*Chen Chali


I iire-Prf one-CL detective.that-CL detective/Charlie Chen
[cP Opi [Akiu hen manyi ei ]].
Akiu very satisfied
'I hired a detective/the detective/Charlie Chen, with whom Akiu is quite
satisfied.'

Note that, given the EMH, the presence of a definite object is supposed to
license the secondary predication. The contrasts in (79) and (80), however,
suggest the opposite. Take (80a) for example:

(81) IP
DPi ... VP
ta ti V'
jiao-guok VP - mapping cycle

subject - DP V' - nuclear scope

yi-ge xuesheng te CP

conglai bu jiao zuoye

166
As illustrated by (81), the clausal predicate conglai bu jiao zuoye'never handed
in homework' may well stand as an inner adjunct in a Larsonian-style structure,
triggering secondary predication on the object DP, just as a non-restrictive
relative does in English. According to the EMH, this should induce specificity
effects on the subject in the current mapping cycle (i.e. the object DP). On the
contrary, the secondary predicate can only cooccur with indefinites such as yi-
ge xuesheng 'a student' in (80a). The same observation applies to yi-ge
zhentan 'a detective' in (80b). To get the complete picture, let's go through the
following table first:

you lai 'come' location S ..a,n veI non-existential


'have' &si 'die' verbs plus Exp/Prf predicates i
ex'stential
assertion of yes yes no no no
object
(in)definiteness
restriction on yes yes no no no
object
allowing
secondary yes yes yes yes no
predication
(inidefiniteness
restriction in the
presence of yes yes yes yes no
secondary
predication

The first row shows that a location verb, unlike "strong" existential predicates
such as you 'have' and lai 'come', conveys primarily the location of its object,
and only secondarily its existence. The situation is thus very similar to that of
possessive you when possession is its primary assertion (cf. the discussion
around (70a,b)). In parallel, an action verb inflected by tle experiential or
perfective aspect denotes primarily an action, and secondarily the existence of
an experience or event. The existence of its object, if any, appears to be an
entailment of the secondary assertion. This pattern is directly related to the
167
contrast in the second row: Only strong existential predicates impose the
(in)definiteness restriction typically associated with there-constructions in
English.
The generalization, however, is only superficial. While the secondary
predication applies only when the existence of object NPs is either asserted or
entailed, as in the third rG;w, its presence reinforces the (in)definiteness
restriction on the part of "weak" existential predicates such as zuo 'sit' and jiao-
guo 'taught', as inthe last row.
As Huang (1987) rightly points out, there are two factors playing around
here: One concerns the (in)definiteness effects induced by e,;stential
predicates, which in turn are conditioned by the presence of secondary
predication when the predicates are less existential; the other is a general
condition on predication, as stated below:

(82) In sentences with a secondary clausal predicate in the sentence-final


position, the NP in the object position must be specific.

The rationale behind (82), as provided by Huang (1987), is that the subject of
continuative description has to be referential. This is exactly the intuition that we
set to capture under the EMH.
As a result, if the (in)definiteness restriction and the predication
requirement (82) apply at the same time, the object NP in question is bound to
be a specific indefinite (i.e., a presuppositional indefinite in Diesing's terms).
Following is the evidence cited by Huang to support this conclusion (with
additions and modifications for ease of exposition):

(83) a. wo jiao-guo yi-ge xuesheng.


I teach-Exp one-CL student
'I have; taught a (certain) student.'
168
b. wo jiao-guo xuesheng.
I teach-Exp student(s)
'I have taught (Sm) student(s).'

(84) a. wo jiao-guo yi-ge xuesheng [cpOpi[ ei hen congming]].


I teach-Exp one-CL student very clever
'I have taught a (certain) student, who is very clever.'

b.* wo jiao-guo xuesheng [cp Opi[ ei hen congming]].


I teach-Exp student(s) very clever
'I have taught (Sm) student(s), who are very clever.

As sho%'n by the contrast between (84a,b), while an indefinite with a numerical


specifier (i.e. yi-ge xuesheng 'a (certain) student') allows an extra CP predicate,
a bare indefinite with a non-generic reading (i.e., xuesheng '(Sm) student(s)')
does not. When the secondary predication is not involved, the asymmetry
disappears accordingly, as in (83a,b). The same observation applies to those
predicates with even stronger existential assertions, as illustrated by the
contrast between (85a,b), as well as that between (86a,b):

(85) a. fangjian-li you yi-ge nOhai [cpOpi [ e hen piaoliang]].


room-inside have one-CL girl very pretty
'In the room, there is a (certain) girl, who is very pretty.'

b.* fangjian-li you nOhai [cp Opi[ ei hen piaoliang]].


room-inside have girl(s) very pretty
'7.In the room, there is/are (Sm) girl(s), who is/are very pretty.'

(86) a. e si-le yi-ge ren [cpOpi[ ei mei dai fangdumianjl]].


die-Prf one-CL person have-not wear gas mask
'(lit.) Died a (certain) person, who did not wear a gas mask.'

b.* e si-le ren [cpOpi [ oi mei dai fangdumianjt]].


die-Prf person(s) have-not wear gas mask
169
'(lit.) Died (Sm) person(s), who did not wear a gas mask.'

As a result, the definiteness effects displayed by (79) and (80) does not
constitute counterexamples to the EMH: On the one hand, names and that-NPs
are ruled out due to the incompatibility between existential assertions and
definite expressions. On the other, specific indefinites are still allowed in all four
types of existential constructions. This guarantees the success of the secondary
predication since the restriction imposed by 3-closure has been met (cf. section
2.1).
What is left unsolved, as noted by Huang in his conclusion, is the exact
relationship between the (in)definiteness restriction and the predication
requirement (82). In other words, problems still arise as to how 3-closure and
secondary predication conspire to reinforce the (in)definiteness restriction, as in
(79) and (80), which are otherwise invisible, as in (77) and (78). It is in this
respect that the VP-shell analysis really shines; As we have sketched in (81),
the domain of the complex predicate (i.e., the secondary clausal predicate) may
well be extended to the trace left by the main verb, and hence to the verb itself
on the assumption that a Chain as a whole is an LF object (cf. Chomksy 1991).
Consequently, the existential assertion (or more precisely, the entailment of the
assertion) of jiao-guo 'teach-Exp' is predicated of the object DP along with
whatever properties the secondary predicate denotes. This results in the
(in)definiteness restriction for the reason just mentioned. In contrast, when
clausal predicates are absent, as in the case of (77) and (78), the existential
entailment remains silent.

170
2.3.3. A Dilemma

Now we have an apparent dilemma at hand. As mentioned above, the


object DP in (81) (i.e., the subject of secondary predication) has to be a specific
indefinite due to the conspiracy between the (in)definiteness restriction and the
EMH. However, the mere existence of specific indefinites casts doubt on
Cheng's (1991) claim that Chinese indefinites are non-quantificational, as long
as we follow Diesing (1992a) in identifying specificity with presuppositionality
(which in turn triggers QR). Then what if we claim instead that Chinese
indefinites are just like their English counterparts, i.e., being ambiguous
between cardinal and presuppositional readings? This move, however,
undermines Cheng's analysis of the specificity/definiteness effects associated
with Chinese subjects (cf. section 1.1). Even if we confine the stipulation to
indefinite objects, bare indefinites still remain uncommitted, as we have seen in
(b) clauses of (84-86).
A possible way out is to say that the presence of specificity in the
indefinites has something to do with existential sentences in general, based on
the fact that the clausal predicates almost exclusively appear in existential
constructions. This possibility is first pointed out by Huang, but then rejected on
the ground that there are verbs which allow the secondary predication, but do
not appear to have anything to do with existence (data from Huang 1987):

(87) a. wo hen xiang xuan yi-men ke [cP Opi [tamen shuo [ei hen youqu]].
I very hope select one-CL class they say very interesting
'Ivery much hope to select a course, which they say is very interesting.'

b. wo zheng zai-kan yi-ben shu [cpOpi[ei hen youyisi]].


I right-now Prg-see one-CL book very interesting
'Right now I am reading a book, which is very interesting.'

171
As shown by (87a,b), verbs like xuan 'select' and kan 'read' do not make
existential assertion. Still, they have no problem with the extra CP predicate.
Nevertheless, they do make existential presuppositions, namely, presupposing
the existence of their objects. In other words, they fall under the category of use-
type verbs in Diesing's (1992) sense. Once we substitute creation verbs like kai
'open' and xie 'write', the sentences degrade:

(88) a.* wo hen xiang kai yi-men ke [Opi [tamen shuo [ ei hen youqu]].
I very hope open one-CL class they say very interesting
'I very much hope to open a course, which they say is very interesting.'

b." wo zheng zai-xie yi-ben shu [Opi [ei hen youyisi]].


I right-now Prg-write one-CL book very interesting
'Right now I am writing a book, which is very interesting.'

The secondary predication in (87a,b) is thus licensed in the same way as


English non-restrictive relativization is licensed when presuppositional some is
present (cf. (57b) and (58b)). Also note that If we twist (88a,b) a little bit further
by supplying the experiential and perfective aspects, grammaticality improves:

(89) a. wo kai-guo yi-men ke [Opi [tamen shuo [ei hen youqu]].


I open-Exp one-CL class they say very interesting
'Ihave the experience of opening a course, which they say is very
interesting.'

b. wo xie-le yi-ben shu [Opi [ei hen youyisi]].


I write-Prf one-CL book very interesting
'I wrote a book, which is very interesting.'

The observation, however, cannot be taken to indicate that all the existential
sentences make presuppositions. As a matter of fact, none of them (except, of
course, those headed by use-type verbs) make presuppositions about the

172
existence of their objects. This point can be illustrated by adding negation on
the sentences containing specific indefinites (i.e., (a) clauses of (84-86)):

(90) a.* wo mei jiao-guo yi-ge xuesheng [cp Opi[ei hen congming]].
I have-not teach-Exp one-CL student very clever
'*I have not taught a student, who is very clever.'

b.* fangjian-li mei-you yi-ge nihai [cp Opi[ e hen piaoliang]].


room-inside not-have one-CL girl very pretty
'*In the room, there is no girl, who is very pretty.'

c.* e mei si yi-ge ren [cpOpi[ei mei dai fangdumianj0]].


have-not die one-CL person have-not ',,ear gas mask
'(lit.) Didn't die a person, who did not wear a gas mask.'

As shown by (90a-c), the secondary predication fails in the presence of


negation. This suggests that the specificity in question comes from existential
assertions or their entailments, since presuppositions as a rule cannot be
falsified by negation (see, for example, Chierchia and McConnell-Ginet 1990,
Heim 1991). Our position is further backed up by the fact that when we take out
the CP predicate, the sentences become acceptable only if the indefinites are
construed as cardinal/existential, behaving rather like a negative polarity item
under the scope of negation:

(91) a. wo mei jiao-guo yi-ge xuesheng.


I have-not teach-Exp one-CL student
'I have not taught anyr a certain student.'

b. fangjian-li mei-you yi-ge nOhai.


room-inside not-have one-CL girl
'In the room, there is nor not a certain girl.'

c. e mel si yi-ge ren.


have-not die one-CL person

173
'(lit.) Didn't die any/* a certain person.'

That is, the specific reading disappears when existential assertions/entailments


are undone. As a result, the indefinites in (91a-c) can only get licensing from 3-
closure (also cf. Cheng 1991). But what role does the negative polarity reading
play in regard tc predication licensing? Why doesn't it license the predication
clauses in (90a-c)? This again leads us back to the cyclicity of mapping. Let's
take (90a) for example:

(92) IP1
DPi ... Neg'

wo Neg (*3x) VP 1

mei t V'
jiao-guok VP2 -- mapping cycle

subject - DP V' - nuclear scope

yi-ge xuesheng(x) tk CP
Opj ... IP2

ej hen congming

First let's put aside the trivial mapping cycle within the secondary CP predicate,
and focus on the VP shells of the main clause. Bottom-up-wise, the first
predicate encountered is the predication CP. Its immediate c-commanding XP is
the lower 'Y"P shell (i.e., VP2), which constitutes an independent mapping cycle
according to the EMH (11). 3-closure may apply on the nruclear scope
corresponding to the CP node, 2 1 but does not apply since there is no unbound
variable within the predicate.

21 Note that 3-closure actually closes off the V' node of the lower VP shell,
since the domain of the predication CP should be extended to the verb trace to
174
On the other hand, we do have a variable introduced by the object
xuesheng 'student' (i.e., the subject of secondary predication) in the relevant
restrictive clause, which is left unbound at the end of this mapping cycle: It
cannot be licensed by 3-closure on the next cycle (cf. section 1.1), nor can it get
licensing from the experiential aspect -guo, since negation undoes the
assertion of the existence of the student-teaching action/event, as well as the
entailment concerning the existence of the object xuesheng 'student'. As a
result, (90a) is correctly ruled out in mapping theoretical terms. As for (91a),
where the predication CP does not occur, there is only one mapping cycle (i.e.,
IP1). T3. variable in question is thus bound byv-clos.", resulting in the
negative polarity (cardinal) reading.
The conclusion, therefore, appears to be that presuppositionality cannot be
equated to specificity, as far as the predication requirement (82) is concerned.
Rather, presuppositional indefinites should be treated as a proper subset of
specific indefinites. If our observation is on the right track, then the specific
reading of indefinites does not necessarily result from inherent quantificational
force, nor does in-situ licensing of indefinites necessarily involve 3-closure. As
we have seen above, licensing may well come from complex predication given
the EMH, which option is not available in a simple IP-VP split. Consequently,
there is no conflict between (82) and Cheng's claim that Chinese indefinites are
non-quantificational, and her analysis of the specificity/definiteness effects on
Chinese subjects is also maintained.

spell out the existential entailment of -guo. Keeping this in mind, we will
continue to refer to the CP node for the ease of exposition.
175
2.3.4. Bare Indefinites

A notable drawback under our approach concerns the bare indefinites in


(b)clauses of (84-86). That is, if it turns out to be true that the specific readings
of (84a-86a) are contributed by secondary predication, it is still unclear why
(84b-86b) with bare indefinites do not benefit from the same application. Here
our hunch is that the licensing between indefinites and existential predicates is
mutual: An indefinite with a numerical determiner like yi-ge 'one-CL' is
intuitively more specific than a bare indefinite, and thus more likely to get
licensing from the existential entailment spelled out by the secondary
predication (cf. section 2.3.2). As a matter of fact, the above intuition conforms to
Carlson's (1977a) observation that there are a number of asymmetries between
English bare plurals (e.g., policemen) and erdinary indefinites (e.g., a
policeman). One of them concerns the lack of specific rsadings on the part of
bare plurals, as shown by the following contrast:

(93) Miles wants to meet a policeman.


a. 3x x is a policeman A Miles wants to meet x
b. Miles wants (3 x x is a policeman A Miles meet x)

(94) Miles wants to meet policemen.


a.#3x x is a policeman A Miles wants to meet x
b. Miles wants (3x x is a policeman A Miles meet x)

(93a) is the "transparent" reaaing in Quine's sense, which can be paraphrased


roughly as 'There is a (particular) policeman that Miles wants to meet'. (93b), on
the other hand, does not refer to any specific policeman, and the indefinite in
question takes the narrow scope. This ambiguity, as Carlson points out, does
not hold for the bare plural policemen in (94). Namely, the transparent/specific
reading is missing.

176
The same thing can be said about Chinese bare indefinites, which can be
either plural or singular:

(95) Akiu xiang zhao yi-ge jingcha.


Akiu want send-for one-CL policeman
a. 3x x is a policeman A Akiu wants to send for x
b. Akiu wants 3x (x is a policeman A Akiu sends for x)

(96) Akiu xiang zhao jingcha.


Akiu want send-for policeman/policemen
a.#3x x is a policeman A Akiu wants to send for x
b. Akiu wants 3x (x is a policeman A Akiu sends for x)

As shown by the above contrast, the bare indefinite jingcht


'policeman/policemen' in (96) cannot refer to a particular policeman, whereas
its more "specified" counterpart in (95) can. At the other end of the scale, jingcha
may also refer to a kind, serving as a proper name, just as its English
counterpart (cf. Carlson 1977a):

(97) Akiu jian-guo jingcha.


Akiu meet-Exp policeman/policemen
'Akiu met policemen (this kind of people) before.'

Nonetheless, differences do exist. Chinese bare indefinites allow a curious


kind of construal which is best described as "diectic" or "demonstrative; ', as
shown by (98):

(98) ren lai-le.


person(s) come-inc
'That/Those person(s) is/are just coming.'

1 77
The closest paraphrase of (98) is literally ta lai-le '(S)he is just coming' or tamen
lai-le 'They are just coming'. This usage is also found in the object pcsition,
typically in alternation with nun-specific construals such as k99b):

(99) Akiu zhao-dao ren le.


Akiu find-reach person(s) Inc
a. Akiu has found that/those person(s).
b. Akiu has found (Sm) person(s).

Again, (99) can be paraphrased as Akiu zhao-dao ta(men) le 'Akiu has found
her/him/them'. For our purpose here, it suffices to recognize that the deictic
reading may well correlate to the fact that Chinese is a pro-drop language,
where ren 'person' can be headed an empty demonstrative.
The generalization then seems to be that the resistence of specificity
construals is an attribute shared by bare (in)definites across languages. In other
word, these "bare DPs" can be either definite (i.e., the deictic and kind
readings), or non-specific (i.e., the generic and existential readings), but never
in between. The deviance of (84b-86b) thus falls under our account: On the one
hand, the definite readings are blocked by the (in)definiteness restriction
imposed by existential predicates. The existential reading, on the other, is
blocked because 3-closure is not available in the mapping cycle of the
secondary predicates.22 As a result, bare indefinites are disallowed wherever
the conspiracy between the (in)definiteness restriction and the EMH takes
effect.

22 Note that bare indefinites generally do not construe as generic in the


object position.
178
2.4. A Refinement

As a reflection, one may wonder, if our conception of the EMH is correct,


why non-restrictive relativization fails where secondary predication is allowed,
as made clear by comparing (100a,b) with (101a,b) respectively ((72b), (76b),
(85a), and (86a) repeated below):

(100) a. fangjian-li you yi-ge n;Ohai [P Opi[ ei hen piaoliang]].


room-inside have one-CL girl very pretty
'In the room, there is a (certain) girl, who is very pretty.'

b. e si-le yi-ge ren [c Opi [ e mei dai fangdumianjl]].


die-Prf one-CL person have-not wear gas mask
'(lit.) Died a (certain) person, who did not wear a gas mask.'

(101) a.* fangjian-li you [DP [CP Op, [ ei dai-zhe hong maodou]] de yi-ge n0hai].
room-inside have wear-Dur red cloak PNM one-CL girl
'. In the room, there is a girl, who wears a red cloak.'

b.* e si-le [DP[CPOPi[ ei bu xi-zao]] de liang-ge ren] le,


die-Prf not take-bath PNM two-CL person Prt
'(lit.) Died two persons, who don't take bathes.'

To put the question in a different way, on what grounds do existential predicates


such as you 'have' and (dis)appearance verbs such as si 'die' discriminate
between non-restrictive relatives and secondary predicates in regard to
predication licensing?
An immediate answer comes from their distinct mapping geometries: While
existential assertions can be made available to indefinite objects by extending
the domain of secondary predicates to verb traces (and hence to the whole Vo-
Chain in (102a)), the same access is denied to the subjects of non-restrictive
relativization (i.e., the head nouns in (101a,b)), because the relevant mapping

179
cycle is the higher DP node in (102b), and there is no way to relate existential
predicates to non-restrictive relatives:

(102) a.
Vk VP - mapping cycle

subject *- DP V' -- nuclear scope

tk CP
Op IP

V'
Vk VP

mapping cycle -- E DP - I

nuclear scope *- CP DP -subject tk


Op IP

The same observation applies to weak existential predicates such as


action verbs inflected by the experiential/perfective aspects, as illustrated by the
contrasts between (103a-c) with (104a-c):

(103) a. wo jiao-guo yi-ge xuesheng [cp Opi [ e hen congming]].


I teach-Exp one-CL student very clever
'I have taught a (certain) student, who is very clever.'

b. wo kai-guo yi-men ke [cP Opi [tamen shuo [ ej hen youqu]].


I open-Exp one-CL class they say very interesting
'I have the experience of opening a course, which they say is viry
interesting.

c. wo xie-le yi-ben shu [cp Opi[ei hen youyisi]l.


I write-Prf one-CL book very interesting
'I wrote a book, which is very interesting.'
180
(104) a." wo jiac-guo [DP [CP Op [ei hen congming]] de yi-ge xuesheng].
I teach-Exp very clever PNM one-CL student
'I have taught a (certain) student, who is very clever.'

b.* wo kai-guo [o.p [Cp Opi [tamen shuo [ ei hen youqu]] de yi-men ke].
I open-Exp they say very interesting PNM one-CL class
'I have the experience of opening a course, which they say is very
interesting.'

c.* wo xie-le oDP C?Opi[e 1 hen youyisi]] de yi-ben shu].


I write-Prf very interesting PNM one-CL book
'1wrote a book, which is very interesting.'

Although both (103) and (104) involve predicaticn rather than modification, the
configuration for predication licensing turns out quite differently due to their
distinct mapping geometries, as we have seen 0,, (102a,b). Once we put the
relative clauses under the scope of numerical specifiers, the contrasts
disappear, as evidenced below:

(105) a. wo jiao-guo [Do yi-ge [cp Opi [ej hen congming]] de xueshleng].
I teach Exp one-CL very clever PNM student
'I have taught a (certain) student who is very clever '

b. wo kai-guo [DP yi-men [cP Opi [tamen shuo [ ei hen youqu]] de ke].
I open-Exp one-CL they say very interesting PNM class
'I have the experience of opening a coure9 which they say is very
interesting.'

c. wo xie-le [OP yi-ben [cp Op [ei h'kri youyisi]] de shu].


I write-Prf one-CL very interesting PNM book
'I wrote a book which is very interesting.'

This is because what is involved in (95a-c) is modiiication. Consequently, the


restrictivc relatives do not constitute independent mapping cycles. Rathsr, they
181
are mapped into the relevant restrictive clauses. The EMH thus correctly
predicts that the predication requirement (83) is irrelevant here.
A noticeable problem with the view presented so far is that, although the
EMH makes the right prediction in the configuration (102b), the semantic type
produced is wrong. As we point out in section 2.2, the higher DP node (i.e., the
curren,' mapping cycle) should correspond to an entity or individual rather than
a proposition. A possible way out, as Ncam Chomsky (p.c.) points out, is to
claim that the one-to-one correspondence between syntactic structures and
semantic representations simply cannot be maintained everywhere. For one
thing, the non-restrictive usage of adjectives obviously needs independent
treatments other than the EMH (cf. the discussion around (52)).
Alternatively, we may reconsider the empirical status of non-restrictive
relatives in Chinese. As a matter fact, Chao (1968) argues for a virtually
opposite view of the restrictive/non-restrictive distinction. According to Chao, an
"inner" relative is non-restrictive (or descriptive in his terms), whereas an "outer"
relative is restrictive. He nevertheless points out that the former could also be
restrictive when contrastively stressed. It should be admitted that at least some
of the speakers consulted have difficulty in deciding whether the outer relative
in (106) is restrictive or not:

(106) Akiu kaichu-le [DP [CP Opi [ ei bu xizao]] de na-gersan-ge ren].


Akiu fire-Pri not bathe PNM that-CL/three-CL person
'Akiu fired those/three people(,) who do not take bath.'

But they all agree that the contrast between na-ge'that-CL' and san-ge 'three-
CL' is clear-cut, and the most natural reading of the inner relative in (107) is
restrictive:

182
(107) Akiu kaichu-le [DP na-ge/san-ge [cp Opi [ ei bu xizao]] de ren].
Akiu fire-Prf that-CL/three-CL not bathe PNM pers)n
'Akiu fired those/three people(,) who do not take bath.'

A classic argument for the non-restrictiveness of inner relatives comes from


Huang (1982). He points out that an inner relative can appear in an appositive
expression, as in (108a), while its outer counterpart cannot, as in (108b):

(108) a. niuyue, [DP zhe-ge [cP Opi [renren dou xiaode ei ]] de chengshi].
New York this-CL everyone all know PNM city
'This city, New York, which everyone knows.'

b. *niuyue, [DP [CP Opi [renren dou xiaode ei Je zne-ge chengshi].


New York everyone all know PNM this-CL city

Our worry is that the relative clause in question is only a part of an appositive,
not the appositive itself. In fact, we would translate (108a) as 'New York, the city
which everyone knows' rather than the one given above. The deviance of
(108b), on the other hand, does indicate that there is something wrong with our
analysis sketched earlier. The real problem, therefore, still lies in the outer
relatives.
As noted by Jim Huang (p.c.), there is a middle ground between the two
opposite views. That is, we may entertain the possibility that all relative clauses
are restrictive in Chinese, and the closest equivalents of English non-restrictive
relatives are actually those secondary predicates in existential constructions.
This move, however, renders the EMH irrelevant for the specificity displayed by
(106). Namely, since the outer relative does not trigger predication, there is no
mapping-theoretical account (such as that given in (102)) available for the
asymmetry between "outer" relativization and secondary predication. Therefore,
we have to find another way to characterize the relatives in question.

183
Carlson (1977b) provides a suggestive clue: He argues that there is a type
of relatives which can neither be classified as restrictive nor as non-restrictive.
One of the defining properties is that their head nouns can only take definite
articles, universal quantifiers, and free-choice any as their determiners, as
illustrated by the following contrasts:

(109) a. The people [there were on the life-raft] died.


b. Every person [there was on the life-raftj died.

(110) a. *Several people [there were on the life-raft] died.


b. *Each person [there was on the life-raft] died.

Carlson calls the type "amount relatives", since the constructions at issue
typically involve cardinality of some sort, as is self-evident in the above there-be
clauses (see also Helm 1987). He proposes that there is an cardinal expression
associated the relativized NP in (109) and (110), which in turn induces
(dis)matching effects on the matrix determiners. For instance, while the forty
men and every ten minutes are well-formed, *several many ladies and *each
fifty minutes are not. The same trait is also found in Chinese outer relative, as
exemplified below:

(111) a. Akiu kaichu-le [DP [CP Opi [ ei bu xizao]] de mei-yi-ge ren].


Akiu fire-Prf not bathe PNM every-one-CL person
'Akiu fired every person who do not take bath.'

b. *Akiu kaichu-le Jp [cp Opi [ ei bu xizao]] de xuduo(-de) ren].


Akiu fire-Prf not bathe PNM several person
'Akiu fired several persons who do not take bath.'

It thus appears that the property of an English relative is determined by the


type of constructions it contains, while that of a Chinese relative is determined

184
by the type of constructions it modifies. It is quite possible that the outer relative
in question is not an adjunct of DP, but an adjunct of so-called "measure
phrase" (MP), a trio consisting of a determiner, a numeral, and a classifier (e.g.,
mei-yi-ge 'every-one-CL' in (111la); see also Huang 1982 and Tang 1990), as
illustrated in (112b):

(112) a. inner relative:


DP
D(=MP) N'
mei-yi-ge CP N'
Op IP ren

b. outer relative:
DP
D(=MP) N'
CP D(=MP) ren
Op IP mei-yi-ge

An inner relative, as in (112a), can never be a part of an MP due to its structural


height. Hence the absence of (dis)matching effects in (107). On the other hand,
the seeming specificity effect of (106), as well as that of (111), can be attributed
to the amount construals of outer relatives.
Furthermore, our analysis is also consistent with the well-known scope
interaction between MPs anr' other prenominal modifiers (cf. Huang 1982). For
example, (113a) and (113b) have exactly the same lexical items. But with a
switch of the word order between the MP and the relative clause, different
interpretations emerge:

(113) a. Akiu kaichu-le [DP na-san-ge [cpOpi [ ei bu xizao]] de ren].


Akiu fire-Prf that-three-CL not bathe PNM person

185
'Akiu fired those three people who do not take bath.'

b. Akiu kaichu-le [DP[CP Opi [ ei bu xizao]] de na-san-ge ren].


Akiu fire-Prf not bathe PNM that-three-CL person
'Akiu fired that group of three people who do not take bath.'

(113a) sports a scenario where there is a group of people who do not bathe,
and Akiu specifically picked up those three and fired them. (113b), on the other
hand, asserts that there are groups of three people, and Akiu picked up the
group characterized by not taking bath and fired its members. This point can be
further illustrated by contrasting the outer relative:

(114) Akiu kaichu-le [DP CP Op [ ei bu xizao]] de na-san-ge ren],


Akiu fire-Prf not bathe PNM that-three-CL person
bu shi [DP[CPOpi [ ei bu chuan yifu]] de na-san-ge ren]
not be not wear clothes PNM that-three-CL person
'Akiu fired that group of three people who do not bathe, not that group of
three who do not dress.'

Moreover, it is impossible to contrast the MP in (113b), either by setting the


determiner zhe 'this' against na 'that', or by setting the numeral wu 'five' against
san 'three':

(115) *Akiu kaichu-le [DP [CP Opi [ ei bu xizao]] de na-san-ge ren],


Akiu fire-Prf not bathe PNM that-three-CL person
bu shi [DP[CP Opi[ei bu xizao]] de zhe-san-ge/ra-wu-ge ren]
not be not bathe PNM this-three-CL/that-five-CL person
'Akiu fired that group of three people who do not take bath, not this group
of three/that group of five.'

By assuming that the outer relative restricts the MP, as sketched in (112b), we
correctly predicts the group-firing reading of (113b) (vs. the individual-firing

186
reading of (113a)). As for (113a), the pattern of contrasting is the other way
around. Namely, only the MP, but not the inner relative, can be contrasted, as
shown below:

(116) a. Akiu kaichu-le [DP na-san-ge [cP Opi [ ei ou xizao]] de ren],


Akiu fire-Prf that-three-CL not bathe PNM person
bu shi [DP zhe-san-ge/?na-wu-ge [cP Opi [ ei bu xizao]] de ren].
not be this-three-CL/that-five-CL not bathe PNM person
'Akiu fired that group of three people who do not take bath, not this
group of three/that group of five.'

b. *Akiu kaichu-le [DP na-san-ge [cp Opi [ ei bu xizao]] de ren],


Akiu fire-Prf that-three-CL not bathe PNM person
bu shi [DP na-san-ge [cp Opi [ ei bu chuan yifu]] de ren].
not be that-three-CL not wear clothes PNM person
'Akiu fired that group of three people who do not bathe, not that group
of three who do not dress.'

This in turn suggests that, given the structural distinction between (112a,b),
Chao's (1968) contrasting crite, ion for the restrictiveness of relatives is actually
a criterion for their scopal height. That is, only the topmost restrictor can be
contrasted in neutral stress environment. Therefore, the inner relative cannot be
contrasted not because it is non-restrictive, but because it can not take scope
over the MP.
In sum, we have demonstrated that the asymmetry between outer
relativization and secondary predication can be captured even if the original
mapping theoretical account does not hold (cf. (102)). The key lies in the notion
"amount relative", and the fact that the MP trio is unusually productive ii
Chinese. Although it remains to be seen how the (dis)matching mechanism can

187
be formulated in a proper way, our solution appears to point to the right
direction.

2.5. Not Unlikely Extensions


2.5. 1. Small clauses

Given our ana!ysis of secondary predicates in Chinese existential


constructions, the most natural question to ask is whether predication in small
clauses also fall under the general scheme of the EMH. The answer appears to
be positive. First let's consider the following contrast:

(1 17) Akiu ate apples.


a. Akiu aWe Sm apples.
b. Akiu ate the kind "apple".

(118) Akiu ate [sc apples unpeeled].


a. #Akiu ate Sm apples unpeeled.
b. Akiu ate the kind "apple" unpeeled.

The bare plural appies in (117) is ambiguous between the existential reading
contributed by 3-closure, as paraphrased as (117a), and the kind reading in
Carlson's (1977a) sense, as paraphrased as (117b), where apples is a proper
name of a species or a kind. As pointed out by Irene Heim (p.c.), the (b)
readings may result from habitual construals. Namely, the bare plural in
question can be licensed by a generic operator associatea with past tense.
Since both alternatives are compatible with our analysis, we will leave the
choie open here.
This ambiguity, however, disappear when the bare plural ;n question is
predicated by a secondary predicate such as unpeeled, as shown by the
contrast between (118a,b). The reason is transparent from the viewpoint of the
188
EMH: As unpeeled triggers predication in (118), it is mapped into the nuclear
scope, and the SC node by definition constitutes an independent mapping
cycle, as sketched be!ow:

(119) ..... V'


V SC -. mapping cycle

subject <- UP Adj -* nuclear scope


I I
apples unpeeled

As a result, the subject of the current cycle cannot be licensed by 3-closure due
on the next cycle, and the existential reading is ruled out. In contrast, when
apples refers to a kind, it is in itself definite. The kind reading thus survives.
The same observation applies to singular indefinites such as an apple in
(120). Here the ambiguity is between the existential reading of (120a) and the
specific reading of (120b):

(120) Akiu ate an apple.


a. Akiu ate Sm apple.
b. Akiu ate a certain apple.

In a way strikingly similar to its bare plural counterpart, the existential reading of
an apple is disallowed when secondary predication applies, as evidenced by
the contrast between (121a,b)y

(121) Akiu ate [sc an apple unpeeled].


a. #Akiu ate Sm apple unpeeled.
b. Akiu ate a certain apple unpeeled.

Given that (121) also assumes the small clause configuration in (119), the
specificity follows straightforwardly. That is,the local subject must be headed by

189
a strong determiner, as in (122), where only specific a(n) can serve this
purpose:

(122) ..... V'

V SC -- mapping cycle

subject *- DP Adj -* nuclear scope

anx apple(x) unpeeled

Otherwise, the mapping crashes, since the variable introduced by apple is


unbound within the SC node.

2.5.2. Secondary temporal predicates

Another relevant fact comes from some peculiar interaction between object
indefinites and secondary temporal piedicates: Frequency adverbials such as
twice and duration adverbials such as two years often have curious effects on
the interpretation of indefinites (cf. Dowty 1972, Carlson 1977a). As shown by
the now familiar contrast between the kind and existential readings of (123) and
(124), twice and for two years seem to have the same theoretical status as a
secondary predicate in smal! clauses (cf. (118)):

(123) Akiu ate apples twice.


a. #AKiu ate Sm apples twice.
b. Akiu ate the kind "apple" twice.

(124) Akiu ate apples for two years.


a. #Akiu ate Sm apples for two years,
b. Akiu ate the kind "apple" for two years.

190
As one might expect, a similar pattern is found in singular indefinites, as
evidenced by (125) and (126). Also note that the situation depicted in (b)
readings is a little odd, but not impossible in appropriate contexts. 23

(125) Akiu ate an apple twice.


a. #Akiu ate Sm fish twice.
b. ?Akiu ate a certain apple twice.

(126) Akiu ate an apple for two years.


a. #Akiu ate Sm apple for two years.
b. ?Akiu ate a certain apple for two years.

As noted by Irene Heim (p.c.), examples su.,v1 as (123-126) are likely


candidates for manping-theoretic accounts: It is generally assumed that twice
and for two years are predicates of implicit event arguments (cf. Kratzer 1989).
The problem is that it is unclear how to characterize the relationship such that
the relevant mapping geometry involves object indefinites.
For one thing, the usual postulated positions for temporal arguments (e.g.,
the Spec of TP) are too high for our purpose here. A plainly syntactic account of

23 Depending on the choice of verbs, the contrast between bare p!urals and
singular indefinites could be very sharp. Take the achievement verb kill for
example:
(i) a. John killed rabbits twice/for two years.
b.??John killed a rabbit twice/for two years.
This is the so-called "differentiated scope" effect. Traditionally, the solution Is
based on the scope interaction between indefinites and temporal adverbials,
and holds that fishes allows a narrow scope existential reading which is
unavailable for a fish.
In the light of Carlson (1977a), we will take the stand that the reading in
question actually refers to a kind, and rabbits takes the wide scope just as a
proper name does. Narrow scope existential readings contributed by 3-Closure
are ruled out equally for the bare plural in (ia) and the singular indefinite in (ib).
The oddness of (ib), as pointed out by Noam Chomsky (p.c.). should be
attributed to our world knowledge that a rabbit cannot be killed twice (also cf.
Carlson 1977a), and the sentence should be fine if a rabbit can be resurrected
from death.
191
tense structures from Stowell (1993) nonetheless provides us important clues.
Loosely adopting Higginbotham (1985), Stowell proposes that N is predicative
in nature and accordingly generated with an external argument (or an index in
Higginbotham's terms), which in turn can be saturated by binding from
determiners like the and every.

(127) DP
D NP
thei/everyi ei N
I
apple

Sinc8 the external argument in question could be temporal-spatial, as in


yesterday's papers and mountain gorilla, a natural extension along our line is to
claim that what twice and for two years actually predicate upon in (123) and
(124) is the temporal argument associated with the bare plural apples, as
sketched below:

(128) IP
DPi ... VP
Akiu ti V'
atek VP - mapping cycle

subject NP V' - nuclear scope

T(ime) apples tk XP
I
twice/for two years

The same analysis carries over to (125) and (126), except that an apple does
have a determiner (i.e., the indefinite article), though not strong enough to act as
a binder when construed as non-specific (i.e., the Sm reading).

192
This approach, though tentative in nature, appears to be a productive way
to look at a number of mysteries around the relation between object indefinites
and temporal predicates. For instance, as observed by Huang (1991) and Tang
(1991), Chinese bare indefinites are also incompatible with postverbal duration
and frequency phrases:

(129) Akiu chi-guo pinqguo liang-ci.


Akiu eat-Exp apple two-time
a. #Akiu ate that/those apple(s) twice.
b. #Akiu ate Sm apple(s) twice.
c.??Akiu ate the kind "apple" twice.

(130) Akiu zhao-le ren yi-ge xiawu


Akiu look-for-Prf person one-CL afternoon
a. #Akiu has looked for that/those person(s) for the whole afternoon.
b. #Akiu has looked for Sm person(s) for the whole afternoon.
c.??Akiu has looked for the humankind for the whole afternoon.

First note that the deictic readings mentioned in section 2.3.4 are inadequate
here, as in (129a) and (130a), probably because they are discourse-oriented
construals, and incompatible with adverbials expressing frequency and
duration. This point can be further illustrated by the fact the sentences improve
considerably when ordinary definites are substituted:

(131) a. Akiu chi-guo na-zhong pingguo liang-ci.


Akiu eat-Exp that-kind apple two-time
'Akiu ate the kind of apple twice.'

b. Akiu zhao-le na-ge ren yi-ge xiawu.


Akiu look-for-Prf that-CL person one-CL afternoon
'Akiu has looked for the person for the whole afternoon.'

193
The existential readings of (129b) and (130b) are also disnllowed, in parallel
with their English counterpabts in (123a) and (124a). This parallel is predicted
by the EMH, given that Larsonian structur's in the genre of (128) are assigned
to (129) and (130) as well:

(132) ..... V'


chi-gUOk/zhao-lek VP - mapping cycle

subject ~ NP V' - nuclear scope


T pingguo/ren tk XP
I
liangci/yigexiawu

One way to improve the existential construal is to reverse thQ word order
between object indefinites and frequency/duration phrases, as exemplified by
(133a,b):

(133) a. Akiu chi-guo liang-ci (de) pingguo.


Akiu eat-Exp two-time PNM apple
'Akiu ate Sm apple(s) twice.'

b. Akiu zhao-le yi-ge xiawu (de) ren.


Akiu look-fir-Prf one-CL afternoon PNM person
'Akiu has looked for Sm person(s) for the whole afternoon.'

This phenomenon is significant in two ways. First, the reverse of word order
means the change of structural hierarchy. It is highly possible that liang-ci
'twice' and yi-ge xiawu'for the whole afternoon' are no longer in a position to
serve as secondary predicates in (133a,b). "other words, they are out of the c-
command domain of pingguo 'apple' and ren 'person' (cf. Rothstein 1983).
Consequently, no predication occurs in the lower VP shell, and hence the lack

194
of specificity effects on the object indefinites. This intturn lends further support to
our predication story for (129) and (130).
Second, in the reverse order, a prenominal modifier marker -de can be
attached to the frequency/duration phrases, as if they are part of the object
indefinites. This "syntax-semantics mismatch" in Huang's (1993b) sense
indicates that there must be some subtle connection between the bare
indefinites and the pseudo-specifiers which enables the reanalysis (or
incorporation the same effect). Since liang-ci 'twice' and yi-ge xiawu 'for the
whole afternoon' are by no means determiners of ordinary breed (i.e., they are
MPs of verbs, in a way of speaking), it would make much more sense to say that
they actually quantify over the proposed temporal argument rather than the
head noun itself, as sketched in (134):

(134) DP

liangcii /yigexiawui (-de) NP


Ti pingguo/ren

Our analysis is thus compatible with both Tang's (1990) position that the
frequency/duration phrases are actually MPs (or classifier phrases), and
Huang's view that the DP in question is an event-denoting gerundive (IP[+N] in
his terms), headed by a empty verb meaning 'do'.
As a result, (133a,b) not only reinforces our argument for the EMH, but also
demonstrates that the view presented in (128) and (132) is more than sheer
speculation.
The kind readings of (129c) and (130c), on the other hand, are possible if
forced, provided that Akiu is, say, a Martian, who rarely ate fishes or met human
before. This type of construal is most natural when we topicalize the bare
indefinites:
195
(135) a. pingguo (a), Akiu chi-guo liang-ci.
apple (Top) Akiu eat-Exp two-time
'Apples, Akiu ate twice.'

b. ren (a), Akiu zhao-le yi-ge xiawu.


person (Top) Akiu look-for-Prf one-CL afternoon
'The humankind, Akiu has searched for the whole afternoon.'

As in English, Chinese singular indefinites disallow existential readings


when predicated by frequency/duration phrases:

(136) Akiu chi-guo yi-ge pingguo liang-ci.


Akiu eat-Exp one-CL apple two-time
a. #Akiu ate Sm apple twice.
b. ?Akiu ate a certain apple twice.

(137) Akiu zhao-le yi-ge ren yi-ge xiawu.


Akiu look-for-Prf one-CL person one-CL afternoon
a. #Akiu has looked for Sm person for the whole afternoon.
b. ?Akiu has looked for a certain person for the whole afternoon.

Despite the claim that Chinese indefinites are non-quantificational, the


sentences are more than marginal if yi-ge pingguo 'an apple' and yi-ge ren 'a
person' are interpreted as specific, as (136b) and (137b). This is reminiscent of
the dilemma we encountered in section 2.3.3. The solution offered there also
carries over in a straightforward manner, since, just like secondary clausal
predicates, frequency/duration phrases co-occur only with experiential and
perfective aspects:

(138) a. Akiu na-zhe yi-ge pingguo.


Akiu take-Dur one-CL apple
'Akiu holds an apple.'

196
b. *Akiu na-zhe yi-ge pingguo liang-ci.
Akiu take-Dur one-CL apple two-time
'*Akiu holds an apple twice.'

(139) a. Akiu zai-zhao yi-ge ren.


Akiu Prg-look-for one-CL person
'Akiu is looking for a person.'

b. *Akiu zai-zhao yi-ge ren yi-ge xiawu.


Akiu Prg-look-for one-CL person one-CL afternoon
'*Akiu is looking for a person for the whole afternoon.'

(1q38) and (139) shows that durative and progressive aspects, unlike their
experiential and perfective counterparts, are not compatible with secondary
temporal predicates. Consequently, the EMH makes the correct prediction that
(136a) and (137a) are ruled out due to the typical specificity effect induced by
secondary predication:

(140) ..... V'


chi-guok/zhao-lek VP - mapping cycle

subject r- DP V' -- nuclear scope


D NP tk XP

yi-ge E N liangci/yigexiawu
pingguo/ren

(136b) and (137b), on the other hand, are ruled in due to the existential
entailment associated with experiential and perfective aspects, spelled out as
part of the property of the complex predicate V', as illustrated above. This
parallel again confirms our working hypothesis that a frequency/duration phrase
are nothing less than a syntactic predicate in mapping-theoretic terms.

197
2.5.3. Resultative Complements

In addition to frequency/duration phrases, Chinese resultative


complements (RCs) also display traits of secondary predicates. As Huang
(1988, 1992) observes, a resultative complement can be treated as an open
sentence with a subject pro, as shown by the following examples :

(141) a. Akiu da-de [na-ge ren]i [Rc proi zhan-bu-qilai].


Akiu beat-Res that-CL person stand-not-up
Akiu beat that person to the extent that (s)he cannot stand up.

b. Akiu ba [na-ge ren]i da-de [nRcproi zhan-bu-qilai].


Akiu BA that-CL person beat-Res stand-not-up
Akiu beat that person so much as to make her/him unable to stand up.

In the spirit of Larson (1988), the RCs of (141a,b) can be placed in the inner
adjunct position, predicating upon the object na-ge ren 'that person' in the Spec
of the lower VP shell, as illustrated below:

(142) a. V'

da-dek VP -- mapping cycle

subject <- [na-ge ren]i V' -- nuclear scope

tk RC
proi

b. ..... V'

ba VP - mapping cycle

subject - [na-ge ren]i V' - nuclear scope

da-de RC
proi

198
The only difference is that the primary predicate da-de 'beat-Res' undergoes V-
to-V movement in (141a), as in (142a), while it remains in-situ in (141b), with the
upper V node filled by ba, which serves as a place holder for subsequent LF
head movement of da-de, as in (142b) (also cf. Tsai 1993b). Alternatively, ba
can also be analyzed as a light verb of some sort, expressing "affectedness". In
either case, secondary predication applies, assigning an extra PATIENT role to
the object according to Huang (1992). This adds to the default "extent" reading
a "victimizing" touch, which is particularly explicit in the presence of ba.
When we substitute bare indefinites for the definte objects in (141a,b), the
familiar ,attern emerges: The existential reading '-bloclk as a rule, as in
(143b) and (144b):

(143) Akiu da-de reni [RC proi zhan-bu-qilai].


Akiu beat-Res person stand-not-up
a. ?Akiu beat that/those person(s) to the extent that (s)he/they cannot
stand up.
b. #Akiu beat Sm person(s) to the extent 'hat (s)he/they cannot stand up.
c.#Akiu beat the humankind to the extent that they cannot stand up.

(144) Akiu ba reni da-de [Rc proi zhan-bu-qilai].


Akiu BA person beat-Res stand-not-up
a. Akiu beat that/those person(s) so much as to make her/him/them
unable to stand up.
b.#Akiu beat Sm person(s) so much as to make her/him/them unable to
stand up.
c.#Akiu beat the humankind so much as to make them unable to
stand up.

The deictic construal somehow improves over its counterpart in (129) and (130),
as in (143a) and (144a). In contrast, the kind reading degrades, probably

199
because even the Martian scenario falls short in making sense out of (143c)
and (144c).
Resultative constructions with singular object indefinites also behave
slightly different, as shown below:

(145) Akiu da-de [yi-ge ren]i [fc proi zhan-bu-qilai].


Akiu beat-Res one-CL person stand-not-up
a. #Akiu beat Sm person to the extent that (s)he cannot stand up.
b. #Akiu beat a certain person to the extent that (s)he cannot stand up.

(146) Akiu ba [yi-ge ren]i da-de [Rn proi zhan-bu-qilai].


Akiu BA one-CL person beat-Res stand-not-up
a. #Akiu beat Sm person so much as to make her/him unable to stand up.
b.#Akiu beat a certain person so much as to make her/him unable to
stand up.

The specific reading, as well as the existential reading, is blocked. This is


because there is no experiential or perfective aspect in this case, and the
connotation of the resultative maker-de, if any, remains obscure. As a result, no
existential assertion/entailment is available for licensing the object indefinite.
Our position is further strengthened by the fact that, unlike secondary clausal
predicates in existential constructions, RCs do not impose (in)definiteness
restriction, as we have seen in (141a,b).

2.5.4. A preverbal-postverbal asymmetry of temporal adjuncts

A significant implication of the original IP-VP split is that not only


arguments but also adjuncts have to observe specificity, as long as they are
nominals and outside VP. This prediction, interestingly enough, turns out to be
true in Chinese, as evidenced by the contrasts of (147) and (148):

200
(147) Akiu zhe-liang-nian/you-liang-nian/*liang-nian qu-le meiguo.
Akiu this-two-year/have-two-year/two-year go-Prf America
Akiu went to America these two years/for two specific years/for two years.'

(148) Akiu zhe-liang-ci/you-liang-nian/*liang-ci zhu 10guan.


Akiu this-two-time/have-two-time/two-time stay hotel
'Akiu stayed in hotel these two times/two specific times/twice.'

As shown above, frequency/duration phrases like liang-nian 'two years' and


liang-ci 'twice' do not occur preverbally without being headed by strong
determiners such as zhe 'this' and you 'have'. As noted by Li & Thompson
(1981), among others, the descriptive generalization appears to be that
preverbal expressions tend to be specific or definite, whereas postverbal
expressions tend to be non-specific and indefinite.
The only problem with this picture in regard to the EMH, as pointed out by
Ken Hale (p.c.) concerns the (non-)specificity restriction on postverbal adjuncts,
as shown by the following examples:

(149) Akiu qu-le meiguo liang-nian*you-liang-ci Pzhe-liang-nian.


Akiu go-Prf America two-year/have-two-time/this-two-year
'Akiu went to America for two year/for two specific years/for these
two years.'

(150) Akiu zhu-guo 10guan liang-ciryou-liang-ci rzhe-liang-ci.


Akiu stay-Exp hotel two-time/have-two-time/this-two-time
'Akiu stayed in hotel twice/two specific times/these two times.'

Here the pattern of contrasts is reversed. Namely, frequency/duration phrases


like liang-nian'two years' and liang-ci 'twice' cannot occur postverbally when
headed by strong determiners such as zhe'this' and you 'have'. Neither the IP-
VP split nor the EMH makes relevant prediction here, and the cause remains
mysterious.

201
In the light of the discussion in section 2.5.2, however, we are able to
address the issue in a more productive way. Given our proposal that liang-nian
and liang-ci should be treated as syntactic predicates, they must contain some
open position so as to qualify as an unsaturated function. Since zhe and you
only serve to "seal" open positions by saturating unbound arguments (cf.
Higginbotham 1985, Stowell 1989, 1993), it is just natural that they should be
ruled out in this type of configuration.

202
3. Chain-formation as a Copying Operation
3.1. A Few Good Questions

Although the EMH works reasonably well in accounting for the specificity
and (in)definiteness effects resulted from the interaction between predication
and quantification, there are three troublesome aspects when we reflect upon
the issue more closely. The first one concerns the availability of lowering as a
way to derive the ambiguity (i.e., specific/generic vs. existential/cardinal) of
English subject indefinites. The second one concerns the unavailability of
lowering in the presence of stative predicates (or individual-level predicates in
Carlson's (1977a) sense). The last aspect is related to the unavailability of
lowering as a way to salvage Chinese subject indefinites.
Let's consider them one by one. It is well-known that English singular
indefinites are ambiguous, as exemplified below:

(151) A dog ran away.


a. A certain dog ran away. (presuppositional/specific)
b. Sm dog ran away. (existential/non-specific)

The reason, as provided by Diesing (1992a), is that the subject a dog can be
lowered back to its original position, i.e., the VP Spec, given the VP-internal
subject hypothesis, as illustrated by (152b):

(152) a. [Ip A dogi ... (*3) [vP ti [v' ran away]]]. (presuppositional/specific)

b. [(IP ... 3x [VP a dog(x) [v' ran away]]]. (existential/non-specific)

While this type of "reconstruction" effect is very common for A'-chains created by
QR, wh-movement, and VP-fronting, it has been widely observed that A-chains
created by NP-movement do not seem to have the same property. This point

203
can be illustrated by the contrast between (153a,b), as well as that between
(154a,b) (data from Huang 1993a):

(153) a. *Itseems to himi that [DP the claim that Johni overslept] is false.
b. [DP the claim that Johni overslept]k seems to himi [tk to be false].

(154) a. The pictures of Johni surprises himi.


b. *Hei is surprised ti by the pictures of Johni.

If the "heavy" DP in (153b) reconstructs to its initial trace, the sentence should
be as bad as (153a) due to Binding Principle C violation. On the other hand, if
the subject pronoun he in (154b) reconstructs to its base position, the sentence
should be as good as (154a). These predictions, however, are not borne out.
We are thus bound to question the nature of the lowering mechanism.
Second, the stage-individual distinction of predicates has generated a
number of controversies since Carlson (1977a). One of them is whether the
asymmetry can be characterized in structural terms (cf. Lasnik & Fiengo 1974).
In the spirit of Kratzer (1989), Diesing (1992a,b, propose a syntactic account
based on the properties of INFL involved, as illustrated by the contrast between
(155) and (156):

(155) Stage-level Predicate:


IP
Spec I'

S t VP
i___-X ___ -.... .
no 0 Spec V'
NP
t Vo

0
204
(156) Individual-level Predicate:
IP
Spec I'
NP
S 10o VP

0 Spec V'
PRO
t Vo
0

The INFL associated with stage-level predicates is claimed to be


"unaccusative", and does not assign an external 0-role, as in (155). In contrast,
the INFL of individual-level predicates does assign an external 0-role, roughly
expressing "have the property x". And the external 0-role of the main predicate
is assigned to a PRO in the VP Spec, as in (156). The subject of individual-level
predicates thus behaves rather like a topic, which never reconstruct for reasons
mentioned in section 1.2.
Our particular worry here is that any predicate can be said to express
"have the property x" in the first place. It is not clear why this should be
structurally realized for stative or individual-level predicates alone. Furthermore,
the asymmetry in question is also found in the object position. For instance, it is
easy to paraphrase stative or individual-level predicates in terms of consistent
properties, evidenced by (157a,b). Sentences with stage-level predicates
behave otherwise, as evidenced by (158a,b):

(157) People love dogs.


a. People are dog-lovers.
b. Dogs are loved by people.

205
(158) People pat dogs.
a. #People are dog-patters.
b. #Dogs are patted by people.

The stage-individual (or state-action) asymmetry is therefore relevant not only


for subject bare plurals, but also for their object counterparts. For Diesing
(1992a), this would require LF scrambling of the object in (157), but not in (158),
which clearly does not follow from the distinction between (155) and (156). It is
therefore worth while having a second thought on the traditional account before
we make the move.
The third issue concerns the fact that Chinese singular indefinites do not
appear in the IP Spec without "marked" licensing (also cf. section 1.1):

(159) a. *yi-zhi gou pao-le.


one-CL dog run-Prf
'A dog ran away.'

b. yi-zhi gou *(neng) pao duo kuai?


one-CL dog can run how fast
'How fast can a dog run?'

c. you yi-zhi gou pao-le.


have one-CL dog run-Prf
'There is a dog that ran away.'

d. na-yi-zhi gou pao-le.


that-one-CL dog run-Prf
'That dog ran away.'

The singular indefinite yi-zhi gou'a dog' is licensed by a modal in (159b), by a


existential predicate in (159c), and by a demonstrative in (159d). Without the
licensing, the sentence is simply ruled out, as in (159a). The same observation
by and large holds for bare indefinites as well:
206
(160) gou pao-le.
dog run-Prf
a. That/Those dog(s) ran away.
b.#Sm dog(s) ran away.

(161) gou yao weiba


dog swing tail
a. Dogs, in general, swing tails.
b. #Sm dogs swing tails.

(162) you-de gou pao-le.


have-PNM dog run-Prf
'Some of the dogs ran away.'

(163) na-xie gou pao-le.


those dog run-Prf
7Those dogs ran away.'

As shown by the contrast between (160a,b), the existential reading is ruled out,
while the deictic reading survives. As for (161), only the generic reading is
possible, presumably induced by the generic tense. (162) and (163) show
patterns similar to (159c,d) respectively. As proposed by Cheng (1991), these
facts will follow from the IP-VP split directly if we assume that Chinese
indefinites, bare or not, are non-quantificational, and more importantly, that
Chinese subjects do not undergo lowering. We thus appear to have got some
hold of how the linguistic variation can be handled in mapping-theoretic terms.
The problem is that we still do not know why the lowering mechanism should

behave this way.


The solutions which we are going to offer are based on the copy theory
developed by Chomsky (1992) and the notion of individual variable conceived
in Heim (1987) and Frampton (1990). We will start with English indefinites first.
207
3.2. Lowering or Copying?

One of the revealing insights of Chomsky (1992) concerns the observation


that the interpretation procedure known as "reconstruction" in the literature can
be implemented in terms of "copying" instead of lowering. The crucial
assumption is that Move-a leaves behind a copy rather than a trace. This copy
deletes at PF, while providing reconstruction materials at LF, as exemplified
below:

(164) PF: [Which book] did John read t?


LF [Which book] did John read [which "ok]?
a. [which book] John read t - Whichx [x is a book] John read x
b. [which] John read [ t book] - WhichF (F(book) AJohn read F)

Whereas the PF deletion applies under identicalness, its LF counterpart is


further restricted by the need to avoid vacuous quantification. As a result, there
are two converging derivations for the above sentence: First, if the whole copy
deletes at LF, the trace count as a DP variable, as in (164a). The answer to the
question could be either the red one or War and Peace. Second, if the head
noun of the moved DP and the determiner of the remaining copy delete at LF,
the trace count as a D variable (or a functional variable), as in (164b). The
answer could be that (book).
First note that there is no theory-internal reason why the copying
mechanism should discriminate between A-chains and A'-chain. 24 To block
reconstruction in sentences like (154b), we would like to claim that
quantificational expressions are in general subject to copying, as imaplicitly

24 As a matter of fact, there is evidence suggesting that A-chains reconstruct


in psych-verb constructions, as far as Principle A is concerned (cf. Belletti and
Rizzi 1988). But see also Pesetsky (1987b) and Mahajan (1990) for alternative
views.
208
assumed in Chomsky (1992), while referential expressions are not.
Alternatively, we may assume that Move-a always leaves copies, and that in a
chain bearing referential dependency, LF deletion must apply downward due to
the recoverability condition, as illustrated in (165b):

(165) LF: Hei is surprised hei by the pictures of Johni.


a. *Hei is surprised t by the pictures of Johni
b. * t is surprised hei by the pictures of Johni

When the copy does delete, as in (165a), the representation is ruled out
according to Principle C. We will leave the choice open here.
The latter approach also provides us an explanation of the absence of
reconstruction effects in (153b), as shown below:

(166) LF: [DP the claim [cP that John overslept]]lk seems to him [ [DP the claim
[CP that John overslept]]k to be false].
a. [DP the claim [cP that Johni overslept]] seems to himi [ t to be false]
b. *[DP the claim] seems to him [[DP t [cP that John overslept]] to be false]
c. *[DP the] seems to himi [ [DP t claim [cP that Johni overslept]] to be false]

There are a number of ways to implement the deletion besides deleting the
whole copy as in (165a). For ease of exposition, let's pick only the major
constituents, i.e., the claim and that John overslept, as illustrated in (165b,c).
(165b) is ruled out because no operator-variable pair is formed. (165c), on the
other hand, violates Principle C. Consequently, there is no other way to
converge the derivation except deleting the whole copy. The possibility of
reconstruction is therefore blocked.
The next step is to examine how copying works for English indefinites.
Deriving the specific reading for a dog in (151) is quite straightforward, as
shown by (167a):
209
(167) LF: [lp a dogk [VP a dogk ran away]].
a. [IP a dog [VP t ran away]] -- 3x [x is a dog] x ran away
b. [IP [vP a dog ran away]] -* 3x (x is a dog Ax ran away)

For the non-specific reading, we do not have a proper operator-variable pair


immediately after upward deletion. However, 3-closure introduced by
subsequent mapping salvages the derivation, as illustrated in (167b). The same
observation applies to bare plural subjects, as shown below:

(168) LF: [Ip dogSk [vP dogSk bark]].


a. [ip dogs [vP t bark]] -* Gen [x is a dog] xbark
b. [IP [VP dogs bark]] -- 3x (x is a dog Ax bark)

An obvious advantage of the copy theory is that we no longer need to


worry about an asymmetry between over wh-extraction and LF quantification.
That is, under the trace-leaving approach, something more has to be said about
why representations such as (169) are allowed at LF, but not at PF:

(169) Whichi did John read [ ti book]?

Traditionally, this fact can be made to follow from the head-government


requirement or the leftness condition in Ross's (1967) sense, but only at the cost
of stipulating that these constraints apply only at S-structure or PF (e.g., WAHL's
(1987) split ECP approach). Now by assuming that PF deletion applies under
identicalness and in accordance with temporal sequence (i.e., irghtward), 25 we
may derive the asymmetry in a straightforward manner (cf. (164)).

25 See also Kayne (1993) for interesting discussion as to how temporal


sequence may restrict the possibiltiy of word orders in UG.
210
3.3. An Individual Variable Account of Stage-Individual Asymmetries

At first glance, it might seem that we are simply reformulating the problem
to avoid controversies around the lowering hypothesis. Below we will show that
there is much more to the copy theory than just an alternative to derive
reconstruction effects.
A natural extension of our view is that tne VP-internal subject hypothesis
should be maintained for both stage-level and individual-level predicates, and
the difference is that the former allows a copy in the VP Spec, thus subject to
reconstruction, whereas the latter do not. The problem, of course, is why this
should be the case.
A suggestive clue comes from Frampton (1990). Based on Heim (1987),
He points out that the referential/non-referential asymmetry of wh-extraction in
Cinque's (1989) sense can be recast as an asymmetry between individual
variables and amount/degree variables (also cf. Cresti 1994). The distinction
roughly corresponds to that of DP and D variables drawn by Chomsky (1992)
(see also Chomsky 1977, Cooper 1983, among others). For example, how
many books can have two types of interpretation, depending on its logical
representations, as illustrated below:

(170) How many books does Bill think that Mary read?
a. (howx) [x many books: y] Bill think that Mary read y
b. (howx) Bill think that Mary read (x many books)

(170a) represents the reading where a set of books is presupposed. The whole
copy is deleted, and the object variable (i.e., y) has the standard interpretation
for wh-traces, counting as an R-expression. As for (170b), no presupposition is
made, and x many books is construed as a function variable (cf. Engdahl 1980,

211
Reinhart 1992,1993). In this case, the highest N' of the moved wh-phrase and
the determiner of the in-situ copy are deleted.
This ambiguity, however, disappears when wh-island constructions are
involved, as evidenced by the contrast between (171 a.b)

(171) How many books does Bill wonder whether Mary read?
a. (howx) [x many books: y] Bill wonders whether Mary read y
b.#(howx) Bill wonders whether Mary read (x many books)

The reason, as offered by Frampton, is that the long-distance dependency in


question can only be licensed by virtue of the variable it dwells upon, as
formulated below:

(172) A trace of long movement must be interpreted as an individual variable.

If the variable refers to an individual, long wh-movement is allowed according to


(172). In contrast, if the variable refers to an amount or a degree, how many
books can only undergo successive cyclic movement, inducing the wh-island
effect of (171b). In copying theoretic terms, this would mean that long movement
always requires deletion of the entire copy.
Now compare the individual-amount asymmetry with the following contrast:

(173) How many people are available?


a. (howx) [x many people:y] y are available
b. (howx) (x many people) are available

(174) How many people are admirable/intelligent?


a. (howx) [x many people:y] y are admirable/intelligent
b. #(howx) (x many people) are available/intelligent

212
(173) contains a stage-level predicate, i.e., available, and allows both the
individual and amount readings. (174), on the other hand, contains an
individual-level predicate, i.e., admirable, and allows only the individual
reading, where a group of people is presupposed. A working hypothesis thus
can be sketched to capture this parallel:

(174) Individual-level predicates can only predicate upon individual variables.

An immediate question coming to mind is whether the terms "individual" from


both parties refer to the same thing. According to Carlson (1977a), an
(individual) object can defined as a collection of a series of stages which
roughly correspond to spatiotemporal slices in terms of intensional semantics.
An individual-level predicate thus expresses properties consistent to all stages
of its subject, or all members of its subject as a group. In the same vein, an
individual variable refers to an individual group (or an individual kind in
Carlson's terms). Furthermore, the amount/degree construal also appears to
have similar semantics as the stage construal, at least in purely formal terms:

(175) a.

S s S S
b.
0 0 0 0 0

al a2 a3 a4 as5

213
As shown by (175a), the stage construal essentially slices an object according
to some arbitrary spatiotemporal coordinates. In comparison, the
amount/degree construal divides a group of objects according to the given
criterion of measurement (e.g., x many books), as illustrated in (175b).
As a result, it follows from (174) that stative or individual-level predicates
cannot predicate upon amounts or degrees, since they have the same
theoretical status as stages. There-be constructions thus provides an ideal
testing ground for our hypothesis, where the amount/degree construal is
obligatory. This intuition has been formulated by Heim (1987) in the following
terms:

(176) *There be x, where x is an individual variable.

In other words, (174) predicts that stative or individual-level predicates is


incompatible with the structural object of there-be constructions. The prediction
is borne out, as evidenced by the contrast between (177a,b):

(177) a. There are people available.


b. *There are people admirable/intelligent.

Further support comes from DPs of artificial measurement like how many
pounds. As observed by Cinque (1989,1990) and Rizzi (1990), long wh-
movement is typically not available for these so-called non-referential DPs,
which, translated in Heim-Frampton's terms, means that they can never leave
individual variables. Here (174) again makes the right prediction: That is, stative
or individual-level predicates cannot predicate upon measurement DPs, as
evidenced by the respective contrasts between (178a,b) and (179a,b):

(178) a. How many pounds are undetectable/unmeasurable (for this scale)?

214
b. How many pounds are beyond detection/measurement (for this
scale)?

(179) a. *How many pounds are unimaginable/unbelievable (for human)?


b. *How many pounds are beyond imagination/belief (for human)?

Undetectable and unmeasurable are, in a sense, "amount-level" or "degree-


level" predicates. Consequently, they are compatible with the amount variable
(i.e., x many pounds), as in (178a). In contrast, since there is no notion such as
"an individual group of pounds", individual-level predicates such as
unimaginable and unbelievable are ruled out in the presence of how many
pounds, as in (179a). The same analysis obtains for the contrast between
(178b) and (179b).
If our observation turns out to be on the right track, then there is a way to
characterize the stage-individual (or state-action) asymmetries in copying-
theoretic terms, that is, in terms of their logical representations rather than their
structural representations such as (155) and (156): A stative or individual-level
predicate requires absolute deletion of the copy in the VP Spec because its
subject can only be an individual variable, as illustrated by (180a):

(180) LF: [IP firemenk [vP firemenk are admirable]].


a. [hP firemen [vP t are admirable]] -* Genx [x is a fireman] x are admirable
b.*[IP [vP firemen are admirable]] -* 3F F(firemen) are admirable

If LF deletion applies upward, as in (180b), what is left behind is a function


variable (or a D variable in Chomsky's terms). Though 3-closure may undo this
vacuous quantification during the subsequent mapping, the variable in question
fails to satisfy the requirement imposed by admirable. We thus correctly predict
that the existential reading is blocked for firemen in (180). In contrast, there is no

215
such restriction associated with stage-level predicates such as available.
Consequently, both the generic and existential readings are licensed in (181):

(181) LF: [ip firemenk [vP firemenk are available]].


a. [ip firemen [vP t are available]] - Genx [x is a fireman] x are available
b. [IP [VP firemen are available]] -* JF F(firemen) are available

In the light of the above discussion, a tentative account can also be


sketched for objects of stative predicates such as love in (157). Following
Chomsky (1992), we would like to assume that the English object dogs
undergors LF movement to the Spec of AGRo for CTse-ch.tking, as illustrated
below:

(182) AGRsP
Spec AGRs'
peoplei AGRs ... AGRoP

Spec AGRo'
dogsk lovej+AGRo VP1
Spec ... VP 2

ti tk V'

The object copy, as well as the subject copy, undergoes obligatory deletion, as
dictated by the stative predicate love. As a result, both people and dogs are
interpreted as generic by virtue of their positions.
To sum up, by integrating the notion of individual variable into the copy
theory, we have achieved two things. First, the VP-internal subject hypothesis is
maintained for both stage-level and individual-level predicates. Second, the
copying mechanism is generalized to capture the stage-individual distinction,
216
which gives us an edge to solve the seeming reconstruction effects on A-chains.
With this fairly explicit theory in mind, we will proceed to examine Chinese
indefinites.

217
3.4. Disagree Chinese Agreement

As mentioned in section 3.1, the most peculiar property of Chinese subject


indefinites is probably that they never reconstruct to benefit from 3-closure. This
topic-like quality might be attributed to the conjecture that there is no genuine
subject in Chinese-type languages, and all the subject-like items are topics
(see, for example, Tsao 1979). The following parameter proposed by Kim
(1991) may also provide some partial answer to our problem:

(183) Every matrix clause in Chinese-type languages has a topic position


that must be filled overtly at S-structure.

Nevertheless, evidence from VP-reconstruction effects strongly suggests that


subjects originate from the VP Spec in both Chinese and English (cf. Huang
1993a). Moreover, as we have demonstrated earlier (cf. (16)), embedded
subjects do not behave differently from their matrix counterparts. And hence the
irrelevance of (183) as a way to derive the specificity in question.
The solution, in our opinion, still lies in the EMH. While our discussion
mainly focuses on secondary predication, it is instructive to note that, for primary
predication, the nuclear scope is also a relative term. For instance, by moving V
to 1,the nuclear scope may well extend to I' for a subject in the IP Spec. To
derive the facts along this line, it would be necessary that Chinese primary
predicates never move beyond the VP Spec, since only in this way can LF
deletion apply freely without changing the semantics of subject indefinites. In
other words, Chinese subject chains as a whole must be always beyond the
scope of 3-closure, as illustrated in (184a). English, in contrast, requires V-to-I
movement at LF, as illustrated in (184b), presumably due to Case-checking
reasons (cf. Chomsky 1992):

218
(184) a. Chinese LF:
IP
Subjk I'
I VP
Subjk (3) V' - nuclear scope
V
b. English Li:
IP
Subjk (3)1' -- nuclear scope
Vi+l VP
Subjk V'
t

Since the English subject copy "submerges" under the nuclear scope, it is
subject to 3-closure when the head of chain deletes. On the other hand, since
the head of the subject chain is above the nuclear scope, it has to be licensed in
a marked way when its copy deletes. Chinese subject indefinites, in contrast,
are never subject to 3-closure, no matter which way LF deletion goes.
Our task, therefore, is to show that Chinese lacks V-to-I movement, both in
over syntax and in the LF component. As observed by Huang(1993c), there is
solid evidence indicating that Chinese verbs pattern with their English
counterparts in not undergoing S-structure movement:

(185) a. Zhangsan bu xihuan Lisi.


Zhangsan not like Lisi
'Zhangsan does not like Lisi.'
b. *Zhangsan xihuan bu Lisi.
Zhangsan like not Lisi

219
(186) a. Zhangsan changchang ma Lisi.
Zhangsan often scold Lisi
'Zhangsan often scolded Lisi.'
b. *Zhangsan ma changchang Lisi.
Zhangsan scold often Lisi

As shown above, verbs can never locate higher than the negative morpheme
bu and sentential adverbial such as changchang 'often' in Chinese. As a matter
of fact, the same observation applies to auxiliary verbs like you 'have' and shi
'be' as well, as evidenced by the following contrasts:

(187) a. Zhangsan mei you kanjian Lisi.


Zhangsan not have see Lisi
'Zhangsan has not seen Lisi.'
b. *Zhangsan you mei kanjian Lisi.
Zhangsan have not seen Lisi
(188) a. Zhangsan bu shi zuotian lai de.
Zhangsan not be yesterday come DE
'It wasn't yesterday that Zhangsan came.'
b. *Zhangsan shi bu zuotian lai de.
Zhangsan be not yesterday come DE

Probably the only fact which can be taken to suggest otherwise is the
position of Chinese aspects. Namely, they appear to be "inflected" on verbs as
some sort of suffixes. Under some early head movement analysis, this would
mean that verbs have been overtly moved to INFL, given that the projection of,
say, AspP is higher than VP. On the other hand, this won't be the necessary
conclusion if we follow Chomsky's (1992) view that verbs are inserted with
inflection, and arguments Case-marking.

220
Furthermore, if we take a step back and look at the historical development
of Chinese verbal elements, we will realize that the majority of so-called aspect
markers actually come from the second half of compound verbs, which in turn
derive from serial verb constructions. For example, compounds like chi-wan
'eat-finish' and chi-dac 'eat-reach' also have perfective or experiential flavor, in
parallel to chi-le 'eat-Prf' and chi-guo 'eat-Exp'. In fact, -le and -guo still have
contemporary verb counterparts, which literally mean 'finish' and 'pass'
respectively (e.g., liao-duan 'finish-cut' and guo nian 'pass (New) Year'). The
same observation applies to the durative aspect -zhe, as in zhuo/zhao di 'touch
ground'. This indicates that modern Chinese aspects used to be and still are
"secondary" verbs. As pointed out by Alec Marantz (p.c.), if there is an AspP
projection in Chinese, it should locate under VP, rather than above VP.
Our observation, of course, does not guarantee that Chinese verbs do not
move in LF. To prove our case, we have to look at its consequences in a wider
context. First let's assume a more articulated IP structure (cf. Emonds 1978,
Pollock 1989). Saying that there is no V-to-1 movement thus roughly means that
there is no AGRP projection, which hosts LF Case-checking when V features
are weak (cf. Chomsky 1992).
Evidence against the existence of AGRsP in Chinese has come a long
way. As shown by (189a), ta-ziji 'him-self' differs from the geriuine long-distance
anaphor ziji 'self' in that it is clause-bound in object position, behaving exactly
the same way as English himself (cf. Tang 1989, Huang & Tang 1989):

(189) a. Akiui renwei [cP Lisi1 hui xuan ta-ziji-ii)J.


Akiu think Lisi will elect him-self
'Akiui thinks that Lisi1 will elect himself-ilr.'

b. Akiu renwei [c, ta-ziji hui dang-xuan].


Akiu think him-self will get-elected
221
'*Akiu thinks that himself will get elected.'

Nevertheless, the parallel does not hold in subject position: the binding domain
of ta-ziji 'himself' in (189b) is the matrix clause despite the fact that the
embedded clause is tensed. A natural account of this subject/object asymmetry
is that Chinese does not have Agr in IP-related projections, which may serve as
an accessible SUBJECT in defining binding domains (cf. Chomsky 1981,
Huang 1983, Aoun 1985,1986). In addition, Huang(1984,1989b) argues for a
typological distinction between Chinese and Romance pro-drop, based on the
lack of AGR on the part of Chinese: Subject pro is licensed through
identification from discourse, rather than agreement with INFL.
In parallel, Tsai (1993b) demonstrates that the lack of Case-resistance
effects and the presence of scope rigidity will follow naturally if we assume that
there is no AGRoP in Chinese. Following is essentially a reproduction of the
arguments provided there. First compare (190a) with (191a). We find that the
main verb zaiyi 'mind' does not allow an intervening preposition like dui, not
only when taking a DP complement but also when taking a question as its
complement. In contrast, its preverbal counterpart in (190b) cannot be licensed
without being Case-marked by dui, patterning with the preverbal DP
complement in (191b):

(190) a. wo hen guanxin (*dui) [Cp Akiu weishenme bu lai].


I very care about Akiu why not come
'I do care why Akiu will not come.'

b. wo I*(dui) [cP Akiu weishenme bu lai]] hen guanxin.


I about Akiu why not come very care

(191) a. wo hen guanxin (*dui) [Dp zhe-jian shi de qiyin].


I very care about this-CL matter of cause
'I do care about the cause of this matter.'
222
b. wo [*(dui) [DP zhe-jian shi de qiyin]] hen guanxin.
I about this-CL matter of cause very care

Unlike English, this parallel between CP and DP with respect to the Case-
Resistence Principle (CRP, Stowell 1981) is not limited to question
complements. It obtains for proposition complements as well, as evidenced by
the respective parallels between (192a,b) and (193a,b): 26

(192) a. wo hen zaiyi (*dui) [cp Akiu bu lai].


I very mind about Akiu not come
'I do mind Akiu not coming.'

b. wo [*(dui) [CP Akiu bu lai]] hen zaiyi.


I about Akiu not come very mind

(193) a. wo hen zaiyi (*dui) [DP zhe-jian shi].


I very mind about this-CL matter
'I do mind about this matter.'

b. wo [*(dui) [DP zhe-jian shi]] hen zaiyi.


I about this-CL matter very mind

Under the Case-checking approach outlined in Chomsky (1992), the CRP


can then be understood as a condition against vacuous agreement in the sense
that, unlike NPs, CPs do not bear agreement features of their own. The function
of English expletive it is thus two-folded: one is to supply features (always
third person singular) so that Spec-head agreement can be achieved; the other
is to serve as a place holder for CPs so that the Case requirement can be
fulfilled at LF. On the other hand, since there is no T-to-Agrs and V-to-Agro

26 See Tsai (1993b) for more evidence based on sentential subjects and
infinitive complements.
223
movement in Chinese, Case-checking never involves agreement features, and
hence the absence of CRP effects.
Another desirable consequence along this line comes from another
significant distinction between Chinese-type and English-type languages, that
is, the rigidity on scope interaction (or the isomorphism in Huang's (1982)
sense). As is well-known, the following Chinese sentence does not have the
ambiguity which its English counterpart has, as illustrated by the contrast
between (194b) and (195b):

(194) mei-ge nOren dou taoyan mou-ge nanren.


every-CL woman all hate some-CL niai
a. For every x, x a woman, for some y, y a man, x hates y.
b. #For some y, y a man, for every x, x a woman, x hates y.

(195) Every woman hates some man.


a. For every x, x a woman, for some y, y a man, x hates y.
b. For some y, y a man, for every x, x a woman, x hates y.

Now consider the following LF representation of (194), where the object adjoins
to V to check its Case feature in absence of the Spec of Agro:

[v' Iv mouge nanrenj [v taoyan]] . tj.. ]]]]]]


(196) [AspP meige nOreni [Aspo [. . [vp ti
every woman some man hate

Since the head of the object chain does not c-command any member of the
subject chain, the (b) clause reading is impossible in violation of the Scope
Principle (197) proposed by Aoun & Li (1989):

(197) A quantifier A has scope over a quantifier B in case A c-commands


a member of chain containing B.

224
Analyses in the same vein can also be sketched by adopting the notion of
"chain scope" advocated in Kim (1991).
Consequently, it seems safe to assume the LF representation (184a) for
Chinese subject indefinites, and thereby provide a mapping-theoretic account
of the specificity in question.

225
4. Concluding Remarks

To justify the extension of the Mapping Hypothesis (1) and its correlation to
3-closure, We have shown that there is a parallel between subject-predicate
and topic-comment constructions with respect to the range of interpretations
associated with indefinites. We also examined constructions involving non-
restrictive relativization and secondary predication. The conclusion is quite
clear. All these cases bear the hallmark of predication, i.e.,
specificity/definiteness effects on relevant subjects, just as the EMH (11)
predicts in the spirit of the original IP-VP split (cf. Diesing 1992a,b).
Furthermore, the asymmetry between non-restrictive relatives and secondary
clausal predicates also lends support to our claim that the EMH is in work. The
interaction between the (in)definiteness restriction and the predication
requirement (82), on the other hand, suggests that the issue of specificity is
more complicated than previously conceived, and should be treated
accordingly with even greater scrutiny.
Despite of the fact that there are still some technical problems lingering
around, the general approach taken in this paper seems to point to the right
direction, not only on conceptual grounds but also on empirical grounds.

226
REFERENCES

Abe, J. (1993) Binding, Conditions and Scrambling without A/A' Distinction,


PhD Dissertation, University of Connecticut.
Abe, J. (1994) "The Proper Characterization of Minimality," Paper presented at
the Workshop on East Asia Linguistics, UC Irvine, January 1994.
Aoun, J. (1985) A Grammar of Anaphora, MIT Press, Cambridge, MA.
Aoun, J. (1986) Generalized Binding, Foris Publications, Dordrecht.
Aoun, J., Hornstein, N., Lightfoot, D.and A. Weinberg (1987) "Two Types of
Locality", Linguistic Inquiry 18: 537-577.
Aoun, J. and Y.-H. A. Li (1989) "Scope and Constituency," Linguistic Inquiry 20,
141-172.
Aoun, J. and Y.-H. A. Li (1993a) "Wh-elements in Situ: Syntax or LF?," Linguistic
Inquiry 24, 199-238.
Aoun, J. and Y.-H. A. Li (1993b) "On some Differences between Chinese and
Japanese Wh--Elements," Linguistic Inquiry 20, 265-272.
Aoun, J. and Y.-H. A. Li (1993c) Syntax of Scope, MIT Press, Cambridge, MA.
Arhens, K. (1994) "Specificity and Wh-in-situ," to appear in the proceedings
of the Chicago Linguistic Society.
Baltin, M. R. (1982) "A Landing Site Theory of Movement Rules," Linguistic
Inquiry 13, 1-38.
Baker, C. L. (1970) "Notes on the Description of English Questions: the Role of
an Abstract Question Morpheme," Foundations of Language 6.2.
Baker, C. L. (1989) English Syntax, MIT Press, Cambridge, MA.
Barwise, J., and R. Cooper (1981) "Generalized Quantifiers and Natural
Language," Linguistics and Philosophy 4, 159-219.
Bobaljik, J. (1994) "Phrase Structure and Predicates: May Day Musings from
Kamchatka," ms., MIT and Kamchatka Institute of Ecology and Natural
Resources.
Bresnan, J. and J. Grimshaw (1978) "The Syntax of Free Relatives in English,"
Linguistic Inquiry 9, 331-191.
Browning, M. (1987) Null Operator Constructions, PhD Dissertation, MIT.
Chao, Y.-R. (1968) A Grammar of Spoken Chinese, University of California
Press, Berkeley and Los Angeles.
Carlson, G. (1977a) Reference to Kinds in English, PhD Dissertation, University
of Massachussets, Amherst.
Carlson, G. (1977b) "Amount Relatives," Language 53, 520-542.
Chierchia, G. and S. McConnell-Ginet (1990) Meaning and Grammar, MIT Press,
Cambridge, MA.
Chierchia, G. (1992a) "Anaphora and Dynamic Binding," Linguisitcs and
Philosophy 15, 111-183. Kluwer Academic Publishers.
Chierchia, G. (1992b) "Individual Level Predicates as Inherent Generics," ms.,
Cornell University.
Cheng, L. L.-S. (1991) On the Typology of Wh-questions, PhD Dissertation, MIT.
Cheng, L. L.-S. (1993) "On Dou-quantification," ms., UC Irvine.
Cheng, L. L.-S. and C.-T. J. Huang (1993) "Two Types of Donkey Sentence,"
Paper presented at the Fifth North American Conference on Chinese
Linguistics, University of Delaware, and at NELS 24.
227
Chiu, B. (1990) "ACase of Quantifier Floating in Mandarin Chinese," Paper
presented at the Second Northeast Conference on Chinese Linguistics,
University of Pennsylvania.
Chiu, B. (1991) "Suo : an Object-Agreement in Mandarin Chinese," in S.
Barbiers et al. (eds.) LCJL3 Proceedings, Leiden, 77-95.
Chomsky, N. (1964) Current Issues in Linguistic Theory, Mouton, The Hague,
Paris.
Chomsky, N. (1965) Aspects of the Theory of Syntax, MIT Press, Cambridge,
MA.
Chomsky, N. (1975) The Logical Structure of Linguistic Theory, Plenum.
Chomsky, N. (1977) "On Wh-Movement," in P.Culicover, T. Wasow, and A.
Akmajan (eds.) Formal Syntax, Academic Press, New York, 71-132.
Chomsky, N. (1980) Rules and Representations, New York, Columbia
University
Press.
Chomsky, N. (1981) Lectures on Government and Binding, Foris Publications,
Dordrecht.
Chomsky, N. (1986a) Knowledge of Language: Its nature, origin, and use,
Praeger, New York.
Chomsky, N. (1986b) Barriers, MIT Press, Cambridge, MA.
Chomsky, N. (1991) "Some Notes on Economy of Derivation and
Representation," in R. Freidin (ed.)Principles and Parameters in
Comparative Grammar, 417-454, MIT Press, Cambridge, MA.
Chomsky, N. (1992) "A Minimalist Program for Linguistic Theory,"MIT
Occasional Paper in Linguistics #1.
Chomsky, N. (1994) "Bare Phrase Structure," MIT Occasional Paper in
Linguistics #5.
Chomsky, N. and H. Lasnik (1991) "Principle and Parameters Theory," to
appear in J. Jacobs, A. van Stechow, W. Sternefeld, and T. Vennemann
(eds.), Syntax: An InternationalHandbook of Contermporary Research,
de Gruyter.
Cinque, G. (1989) "On the Scope of Long and Successive Cyclic Movement,"
Paper presented at Second Princeton Workshop on Comparative
Grammar.
Cinque, G. (1990) Types of A'-Dependencies, MIT Press, Cambridge, MA.
Cole, P. (1987) "The Structure of Internally Headed Relative Clauses," Natural
Language and Linguistic Theory 5, 277-302.
Cresti, D. (1994) "Extraction and Reconstruction," to appear in Natural
Language Semantics.
Cooper, R. (1983) Quantification and Syntactic Theory, Reidel, Dordrecht.
Cinque, G. (1993) "A Null Theory of Phrase and Compound Stress," Linguistic
Inquiry, 24, 239-297.
Culicover, P. (1991) "Polarity, Inversion, and Focus in English," in Proceedings
of ESCOL 1991, 46-68, Department of Linguistics, The Ohio State
University.
Davidson, D. (1967) "The Logical Form of Action Sentences," in N. Rescher
(ed.) The Logic of Decision and Action, 81-95, University of Pittsburgh
Press, Pittsburgh, PA.
Diesing, M. (1992a) Indefinites, MIT Press, Cambridge, MA.

228
Diesing, M. (1992b) "Bare Plural Subjects and the Derivation of Logical
Representations," Linguistic Inquiry 23, 353-380.
Dowty, D. (1972) Studies in the Logic of Verb Aspect and Time Reference in
English, PhD Dissertation, University of Texas at Austin.
Emonds, J. (1976) A Transformational Approach to English Syntax, Academic
Press, New York.
Emonds, J. (1978) "The Verbal Complex V'-V in French," Linguistic Inquiry9,
151-175.
Frampton, J. (1990) "The Fine Structure of Wh-Movement and the Proper
Formulation of the ECP," ms., Northeastern University.
Fujita, K. (1994) "Middle, Ergative & Passive in English - A Minimalist
Perspective," ms., Osaka Univesity.
Fukui, N.and M.Speas (1986) "Specifiers and Projection," in N. Fukui, T.
Rappoport, and E. Sagey (eds.) MITWPL 8, 128-172.
Fukui, N. (1993) "Parameters and Optionality," Linguistic Inquiry 24, 399-420.
Grimshaw, J. (1979) "Complement Selection and the Lexicon", Linguistic
Inquiry 10, 279-326.
Grimshaw, J. (1990) Argument Structure, MIT Press, Cambridge, MA.
Grimshaw, J. (1993) "Minimal Projection, Heads, and Optimality," ms., Rutgers
University.
Gundel, J. (1974) The Role of Topic and Comment in Linguistic Theory, PhD
Dissertation, Indiana University Linguistics Club, Bloomington, IN.
Hale K. and S. J. Keyser (1991) "On Argument Structure and the Lexical
Expression of Syntactic Relation." ms., MIT.
Heim, I. (1982) The Semantics of Definite and Indefinite Noun Phrases, PhD
Dissertation, University of Massachussets, Amherst.
Heim, 1.(1987) "Where does the Definiteness Restriction Apply? Evidence
from the Definiteness of Variables," in E, Reuland & A. ter Meulen (eds.)
The Representation of (ln)definiteness,21-42, MIT Press, Cambridge, MA.
Heim, I. (1991) "Articles and Definiteness," in A. von Stechow & D.Wunderlich
(eds.) Semantics. An International Handbook of Contemporary Research,
de Gruyter, Berlin.
Higginbotham, J. (1983) "Logical Form, Binding, and Nominals," Linguistic
Inquiry 14, 395-420.
Higginbotham, J. (1985) "On Semantics," LinguisticInquiry 16, 547-593.
Higginbotham, J. (1987) "Indefiniteness and Predication," in E, Reuland &A.ter
Meulen (eds.)The Representation of (In)definiteness, 43-70, MIT Press,
Cambridge, MA.
Higginbotham, J. and R. May (1981) "Questions, Quantifiers, and Crossing,"
The Linguistic Review 1, 41-79.
Hoji, H. (1985) Logical Form Constraints and ConfigurationalStructures in
Japanese, PhD Dissertation, University of Washington, WA.
Horn, L. (1988) "A Presuppositional Analysis of Only and Even," ms., UCLA.
Hou, J. and C. Kitagawa (1987) "Null Operators and the Status of Empty
Categories in Chinese," LinguisticInquiry 18, 518-523.
Hsieh, M.-L. (1994) "On NPI (Negative Polarity Items) Licensing in Chinese,"
ms., USC.
Huang, C.-T. J. (1982) Logical Relations in Chinese and the Theory of
Grammar, PhD Dissertation, MIT.

229
Huang, C.-T. J. (1984) "On the Distribution and Reference of Empty Pronouns,"
Linguistic Inquiry 15, 531-574.
Huang, C.-T. J. (1987) "Existential Sentences in Chinese and (In)definiteness,"
in E, Reuland & A. ter Meulen (eds.) The Representation of
(In)definiteness, 226-253, MIT Press, Cambridge, MA.
Huang, C.-T. J. (1988) "wo pao de kual and Chinese Phrase Structure,"
Language 64, 274-311.
Huang, C.-T. J. (1989a) "Shuo shi he you: jianta zhongwen de dongci fenlei [A
Discussion of 'be' and 'have': with verb classification in Chinese]," ms.,
Cornell University.
Huang, C.-T. J. (1989b) "Pro-Drop in Chinese: A Generalized Control Theory,"
in O.Jaeggli and K. Safir (eds.) The Null Subject Parameter, 185-214.
Huang, C.-T. J. (1991) "More on Chinese Word Order and Parametric Theory,"
ms., UC Irvine.
Huang, C.-T. J. (1992) "Complex Predicates in Generalized Control," in Larson,
R. K., S. latridou, U. Lahiri and J. Higginbotham (eds.) Control and
Grammar, Kluwer Academic Publishers.
Huang, C.-T. J. (1993a) "Reconstruction and the Structure u,VP: Some
Theoretical Consequences," Linguistic Inquiry 24, 103-138.
Huang, C.-T. J. (1993b) "On Lexical Structure and Syntactic Projection," ms.,
UC Irvine.
Huang, C.-T. J. (1993c) "Verb Movement and Some Syntax-Semantics
Mismatches inChinese," ms., UC Irvine.
loup, G. (1975) The Treatment of Quantifier Scope in a Transformational
Grammar, PhD Dissertation, CUNY Graduate Center.
latridou S. and D. Embick (1994) "Conditional Inversion," to appear in the
proceedings of NELS 24.
Jackendoff, Ray (1977) X' Syntax: A Study of Phrase Structure, MIT Press,
Cambridge, MA.
Jonas, D. and J. Bobaljik (1993) "The Role of TP in Icelandic," in Bobaljik,
Jonathan and Colin Phillips (eds.), Papers on Case and Agreement I,
MITWPL 18, MIT, Cambridge.
Jong, F. de and H. Verkuyl (1985) "Generalized Quantifiers: The Properness of
Their Strength," in J. van Benthem and A. ter Meulen (eds.) Generalized
Quantifiers in Natural Languages, 21-43, Foris Publications, Dordrecht.
Kamp, J. A. W. (1981) "A Theory of Truth and Semantic Representation," InJ.
Groenendijk, T. Janssen, and M. Stokhof (eds.) Formal Methods in the
Study of Language, 277-321, Mathematical Centre, Amsterdam.
Kayne, R.(1981) "ECP Extensions," Linguistic Inquiry 12, 93-133.
Kayne, R. (1993) "The Antisymmetry of Syntax," ms., Graduate Center, CUNY.
Katz, J. and P. Postal (1964) An Intergrated Thoery of Linguistic Descriptions,
MIT Press.
Kim, S.(1991) Chain Scope and Quantification Structure, PhD Dissertation,
Brandeis University.
Klima, E. (1964) "Negation in English," in Jerry A. Fodor and Herrold Katz (eds.)
The Structure of Language, Prentice Hall, Englewood, NJ.
Kitagawa, C. (1982) "Topic Constructions in Japanese," Lingua 57, 175-214.
Koizumi, M. (1994a) "Layered Specifiers," to appear in the proceedings of
NELS 24.

230
Koopman, H. and D. Sportiche (1989) "Subjects," ms., UCLA.
Kratzer, A. (1989) "Stage and Individual Levc. Predicates," in Papers on
Quantification, NSF Grant Report, Department of Linguistics, University of
Massachussets, Amherst.
Kroch, A. (1989) "Amount Quantification, Referentiality, and Long Wh-
movement," ms., University of Pennsylvania.
Kuroda, S.-Y. (1965) Generative Grammatical Studies in the Japanese
Language, PhD dissertation, MIT.
Kuroda, S.-Y. (1988) "Whether We Agree or not," Linguisticae Investigationes
12, 1-47.
Larson, R. (1988) "On the Double Object Construction," Linguistic Inquiry 19.3,
335-392.
Lasnik, H. (1993) "Lectures on Minimalist Syntax," UCWPL Occasional Papers
#1.
Lasnik, H. and R. Fiengo (1974) "Complement Object Deletion," Linguistic
Inquiry 15, 235-289.
Lasnik, H. and M. Saito (1984) "On the Nature of Proper Government,"
Linguistic inquiry 15: 235-289.
Lasnik, H. and M. Saito (1992) Move a, MIT Press, Cambridge, MA.
Labnik, H. and T. Stowell (1991) "Weakest Crossover," Linguistic Inquiry 22,
687-720.
Lewis, D. (1975) "Adverbs of Quantification," in E. Keenan (ed.) Formal
Semantics of Natural Language, 3-15, Cambridge University Press,
Cambridge.
Lee, T. (1986) Studies on Quantification in Chinese, PhD Dissertation, UCLA.
Li, Y.-H. A. (1990) Order and Constituency in Manadarin Chinese, Kluwer,
Dordrecht.
Li, Y.-H. A. (1992) "Indefinite Wh in Mandarin Chinese," Journal of East Asian
Linguistics 1, 125-155.
Li, C. and S. Thompson (1981) Mandarin Chinese: A Functional Reference
Grammar, University of California Press, Berkeley.
Lin, J.-W. (1992) "The Syntax of Zenmeyang 'How' and Weishenme 'Why' in
Mandarin Chinese," Journal of East Asian Linguistics 1, 293-331.
Liu, F.-H. (1993) "When does Reciprocity Arise?" ms., University of Arizona.
Liu, F.-H. (1994) "Wh-words in Chinese and in particular," ms., University of
Arizona.
Napoli, D. J. (1989) Predication Theory, Cambridge University Press,
Cambridge.
Mahajan, A. (1990) The A/A'-Distinction and Movement Theory, PhD
Dissertation, MIT.
Mahajan, A. (1993) "On Certain Differences between Hindi and Japanese
Wh-movement," Talk given at the Department of Linguistics, Harvard
University.
May, R. (1985) Logical Form, MIT Press, Cambridge.
Mei, K. (1972) Studies in the Transformational Grammar of Modern Standard
Chinese, PhD Dissertation, Harvard University.
Mei, K.(1978) "Guoyu Yufa zhong de Dongcizu Buyu [On Verb-Phrase
Complements in Chinese]," in A Festschrift for Professor Chu Wan-Li,
511-536, Lianjing Publisher, Taipei.
231
Milsark, G. (1974) Existential Sentences in English, PhD Dissertation, MIT.
Miyagawa, S. (1994) "Nonconfigurationality within a Configurational Structure,"
ms., MIT.
Nishigauchi, T. (1986) Quantification in Syntax, PhD Dissertation, University of
Massachusetts, Amherst.
Nishigauchi, T. (1990) Quantification in the Theory of Grammar, Kluwer
Academic Publishers, Dordrecht.
Oishi, M. (1993) "LF Legitimacy and Chain Formation," ms., Tohoku Gakuin
University, Sendai, Japan.
Parsons, T. (1990) Events in the Semantics of English: A Study in Subatomic
Semantics, MIT Press, Cambridge, MA.
Percus, O.(1994) "Mapping Indefinites," ms., MIT.
Pesetsky, D. (1982) Path and Categories, PhD Dissertation, MIT.
Pesetsky, D. (1987a) "Wh in situ: Movement and Unselective Binding," in E.
Reuland and A. Ter Meulen (eds.) Representation of (In)definiteness, MIT
Press, Cambridge, MA.
Pesetskv. D. (1987b) "Binding Problems with Experiencer Verbs," Linguistic
Inquiry 18, 126-140.
Partee, B. H.(1988) "Many Quantifiers," in Proceedings of ESCOL 1988,
Department of Linguistics, The Ohio State University.
Pollock, J.-Y. (1989) "Verb Movement, Universal Grammar, and the Structure of
IP," Linguistic Inquiry 20, 365-424.
Reuland, E. (1983) "The Extended Projection Principle and the Definiteness
Effect," inM. Barlow, D. Flickinger, and M.Wescoat (eds.) Proceedings of
the West Coast Conference on Formal Linguistics 2, Stanford Linguistics
Association, Department of Linguistics, Stanford University.
Reinhart, T. (1983) Anaphora and Semantic Interpretation,Croom Helm,
London.
Reinhart, T. (1987) "Specifier and Operator Binding," in E, Reuland &A. ter
Meulen (eds.) The Representation of (In)definiteness, 130-167, MIT Press,
Cambridge, MA.
Reinhart, T. (1992) "Interpreting Wh-in-situ," to appear in the Proceedings of the
Amsterdam Colloquim.
Reinhart, T. (1993) "wh-in-situ in the Framework of the Minimalist Program,"
lecture given at Utrecht Linguistics Colloquium.
Rizzi, L. (1986) "On Chain Formation," in H. Borer (eds.) Syntax and Semantics
19: the Syntax of Pronominal Clitics, Academic Press, New York.
Rizzi, L. (1990) Relativized Minimality, MIT Press, Cambridge, MA.
Rizzi, L. (1991) "Residual Verb Second and the Wh-Criterion," ms., Universite
de Geneve.
Rooth, M.(1985) Association with Focus, PhD Dissertation, University of
Massachussets, Amherst.
Ross, J. (1967) Constraints on Variables in Syntax, PhD Dissertation, MIT.
Rothstein, S.(1983) The Syntactic Form of Predication, PhD Dissertation, MIT.
Rudin, C. (1988) "On Multiple Questions and Multiple WH Fronting," Natural
Language & Linguistic Theory 6, 445-502.
Safir, K. (1982) Syntactic Chains and the Definiteness Effect, PhD Dissertation,
MIT.
Safir, K. (1985) Syntactic Chains, Cambridge University Press, Cambridge.

232
Safir, K. (1987) "What Explains the Defi.nteness Effect?" in E,Reuland & A. ter
Meulen (eds.) The Representation of (In)definiteness, 71-97, MIT Press,
Cambridge, MA.
Saito, M. (1989) "Scrambling as Semantically Vacuous A'-movement," in M. R.
Baltin and A. S. Kroch (eds.) Alternative Conceptions of Phrase Structure,
The University of Chicago Press, 182-200.
Stowell, T. (1981) Origins of Phrase Structures, PhD Dissertation, MIT.
Stowell, T. (1989) "Subjects, Specifiers, and X-bar Theory," in M. R.Baltin and
A. S. Kroch (eds.) Alternative Conceptions of Phrase Structure, The
University of Chicago Press, 232-262.
Stowell, K. (1993) "Syntax of Tense," ms., UCLA.
Srivastav, V. (1991) WH Dependencies in Hindi and the Theory of Grammar,
PhD Dissertation, Cornell University.
Srivastav Dayal, V. (1993) "Scope Marking: Its Syntax and Semantics," Paper
presented at MIT Linguistics Colloquium.
Tai, H.-Y. J. (1973) "A Derivational Constraint on Adverbial Placement in
Mandarin Chinese," Journal of Chinese Linguistics 1, 397-413.
Tai, H.-Y. J. (1981) "Temporal Sequence of Chinese Word Order," ms., Southern
Illinois University, Carbondale.
Takano, Y.(1994a) "Scrambling, Relativized Minimlaity, and Economy of
Derivation," Paper presented at WCCFL XIII.
Takano, Y. (1994b) "Predicate Fronting and Internal Subject," ms., UC Irvine.
Tang, C.-C. J. (1990) Chinese Phrase Structure and the Extended X'-Theory,
PhD Dissertation, Cornell University.
Tang, C.-C. J. (1991) "Conditions on the Distribution of Postverbal Duration and
Frequency Phrases in Chinese Revisited," paper presented at the Third
InternationalSymposium on Chinese Languages and Linguistics.
Tang, C.-C. J. (1993) "Adjunct Licensing, Wh-in-situ and the Checking Theory in
Chinese," paper presented at ICCL2, Paris.
Tang, T.-C. (1975) "Contrastive Studies of Chinese and English Relativization,"
The Concentric 1975, 38-66.
Tang, T.-C. (1989) Studies on Chinese Morphology and Syntax: 2, Student
Book Co., Taipei.
Tsai, W.-T. D.(1992a) "On the Absence of Island Effects," Paper presented at
the Fourth North American Conference on Chinese Linguistics, University
of Michigan, Ann Arber.
Tsai, W.-T. D. (1992b) "On Parametrizing Unselective Binding Effects," Paper
presented at the Workshop on East Asia Linguistics, UC Irvine, June 1992.
Tsai, W.-T. D. (1992c) "A Note on Zenmeyang Family," Paper presented at The
3rd International Symposium on Chinese Languages and Linguistics,
National Tsing Hua University, Taiwan.
Tsai, W.-T. D. (1993a) "A Note on Wh-adjunct Assymmetries," to appear in
Chinese Languages and Linguistics, Vol.3.
Tsai, W.-T. D. (1993b) "Visibility, Complement Selection and the Case
Requirement of CP," in J. D. Bobaljik & Colin Philips (eds.) Papers on Case
& Agreement I, MITWPL 18, 215-242.
Tsai, W.-T. D. (1994) "On Nominal Islands and LF Extraction in Chinese,"
Natural Language and Linguistic Theory 12, 121-175.
Tsao, F. (1979) A Functional Study of Topic in Chinese: The First Step Towards

233
Discourse Analysis, Student Book Co., Taipei, Taiwan.
Wang, W. (1965) '"Two Aspect Markers in Mandarin," Language 41.3.
Ura, H. (1994) "Super-raising and the Features-based X-bar Theory," ms., MIT.
Watanabe, A. (1991) "'Wh-in-situ, Subjacency, and Chain Formation," ms., MIT.
Watanabe, A. (1993) AGR-Based Case Theory and Its Interation with the A-bar
System, PhD Dissertation, MIT.
Williams, E. (1980) "Predication," Linguistic Inquiry 11, 203-238.
Yu, X.-L. (1965) "Yiwen Daici de Renzhi Yongfa," Zhongguo Yuwen 1, 30-35.

234